Browse

Journals By Subject

Life Sciences, Agriculture & Food
Chemistry
Medicine & Health
Materials Science
Mathematics & Physics
Electrical & Computer Science
Earth, Energy & Environment
Architecture & Civil Engineering
Education
Economics & Management
Humanities & Social Sciences
Multidisciplinary

Books

Publish Conference Abstract Books
Upcoming Conference Abstract Books
Published Conference Abstract Books
Publish Your Books
Upcoming Books
Published Books

Proceedings

Events

Events
Five Types of Conference Publications

Join

Join as Editor-in-Chief
Join as Senior Editor
Join as Editorial Board Member
Become a Reviewer

Information

For Authors
For Reviewers
For Editors
For Conference Organizers
For Librarians
Article Processing Charges
Special Issue Guidelines
Editorial Process
Manuscript Transfers
Peer Review at SciencePG
Open Access
Copyright and License
Ethical Guidelines
Submit an Article
Research Article | | Peer-Reviewed

AutoMalariaNet: A VGG16-Based Deep Learning Model for High-Performance Automated Malaria Parasite Detection in Blood Smear Images

Emmanuel Osaze Oshoiribhor,  Adetokunbo MacGregor John-Otumu*
Published in American Journal of Neural Networks and Applications (Volume 11, Issue 1)
Received: 20 March 2025     Accepted: 27 March 2025     Published: 17 April 2025
Views:       Downloads:
Download PDF
Abstract

This research paper presents an automated malaria detection system using deep learning techniques to enhance diagnostic accuracy and efficiency, addressing the critical challenge of early and precise malaria diagnosis, especially in resource-constrained regions. Malaria remains a significant global health burden, particularly in tropical and subtropical regions where timely and accurate diagnosis is crucial for effective treatment and control. Traditional diagnostic methods, such as microscopic examination of blood smears, require skilled parasitologists and are often labor-intensive and time-consuming, making rapid detection difficult. To overcome these limitations, this study develops a deep learning-based malaria detection system integrating a Custom Convolutional Neural Network (CNN) and a pre-trained VGG16 model, trained on a publicly available malaria blood smear image dataset from Kaggle. Several data preprocessing techniques, including normalization and augmentation (rotation, flipping, scaling, and brightness adjustment), were applied to improve model generalization and robustness. The system is deployed through a web-based interface developed using Python, Flask, and HTML, allowing users to upload blood smear images and obtain real-time diagnostic results. Experimental evaluations demonstrate that the VGG16 model outperforms the Custom CNN, achieving an accuracy of 97%, precision of 96%, recall of 96.56%, and an F1-score of 97%, whereas the Custom CNN attained an accuracy of 87%, precision of 86%, recall of 85%, and an F1-score of 84.45%. These findings validate the effectiveness of deep learning in automating malaria detection and reducing reliance on manual microscopic examination, offering a scalable and accessible diagnostic tool for healthcare facilities with limited resources. Despite the success of the proposed system, further research is necessary to enhance model interpretability and trustworthiness. Future work should explore the integration of Vision Transformers (ViTs), Large Language Models (LLMs), and Ensemble Deep Learning techniques to improve malaria detection performance. Additionally, Explainable AI (XAI) methods, such as Grad-CAM, should be incorporated to provide visual explanations of model predictions, ensuring transparency and aiding medical professionals in understanding the decision-making process. By integrating these advancements, future systems can enhance both diagnostic accuracy and interpretability, making AI-driven malaria detection more reliable and widely applicable.

Published in American Journal of Neural Networks and Applications (Volume 11, Issue 1)
DOI 10.11648/j.ajnna.20251101.12
Page(s) 11-27
Creative Commons

This is an Open Access article, distributed under the terms of the Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution and reproduction in any medium or format, provided the original work is properly cited.

Copyright

Copyright © The Author(s), 2025. Published by Science Publishing Group
Previous article
Keywords

Deep Learning, Pre-trained, VGG-16, CNN, Malaria, Classification, Blood Smear Images

1. Introduction
Malaria is a serious disease spread through the bite of an infected female Anopheles mosquito carrying the Plasmodium parasite. This infection is common in tropical regions and affects millions of people worldwide
[1, 2]
. Once the parasite enters the bloodstream, it travels to the liver, multiplies, and then spreads to red blood cells. Symptoms of malaria include fever, fatigue, headaches, and, in severe cases, seizures or coma
[3]
. Among the different species of Plasmodium, P. falciparum is the most dangerous and is responsible for most malaria-related deaths, particularly in sub-Saharan Africa
[4]
.
This parasite spreads quickly and can cause severe anemia, organ failure, and cerebral malaria, a life-threatening condition affecting the brain. Malaria remains a major public health issue, especially in regions with high transmission rates. According to the Centers for Disease Control and Prevention (CDC), in 2022 alone, over 608,000 people died from malaria, with young children in sub-Saharan Africa being the most affected
[5]
. The disease persists in areas where access to healthcare, preventive measures, and proper housing is limited
[6]
.
In Nigeria, malaria is one of the leading causes of death, responsible for approximately 300,000 deaths annually. Around half of the population experiences at least one malaria episode per year, leading to significant health and economic burdens. Individuals and families often bear the cost of treatment, travel to healthcare facilities, and preventive measures such as insecticide-treated bed nets. Additionally, the disease contributes to lost work productivity, school absenteeism, and increased government spending on healthcare services
[7]
.
Figure 1 illustrates the malaria transmission cycle, which involves both humans and mosquitoes. The cycle begins when an infected Anopheles mosquito bites a person, introducing Plasmodium parasites (sporozoites) into the bloodstream. These sporozoites travel to the liver, where they multiply and develop into merozoites. Once matured, the merozoites enter red blood cells, continuing to multiply and eventually forming gametocytes.
When another mosquito bites the infected person, it ingests the gametocytes, which mature into sporozoites inside the mosquito’s body. These sporozoites then migrate to the mosquito’s salivary glands, making it capable of transmitting malaria when it bites another person. This cycle repeats, spreading the disease.
Figure 1. Malaria transmission cycle.
Malaria symptoms, including fever and chills, occur when the parasites destroy red blood cells. Effective prevention strategies, such as insecticide-treated nets, antimalarial medications, and mosquito control methods, help reduce transmission and protect communities from infection.
The financial impact of malaria extends beyond healthcare expenses, as the disease negatively affects economic activities and tourism. In highly affected countries like Nigeria, malaria reduces economic growth by an estimated 13% each year
[8]
. Limited healthcare funding, high out-of-pocket expenses, and poor access to medical facilities contribute to the persistence of malaria. Moreover, social and economic inequalities make it difficult for low-income populations to afford preventive and treatment measures. Meanwhile, wealthier individuals often have access to better healthcare services both locally and abroad. Addressing these disparities is crucial for effective malaria control and eradication.
Early and accurate diagnosis of malaria is essential for reducing its impact. The most common method of diagnosis is microscopic examination of blood samples, but this approach requires skilled personnel and adequate facilities, which are often lacking in malaria-endemic regions
[9]
. Given these challenges, automating the malaria detection process can significantly improve diagnosis and treatment outcomes
[10]
.
Technological advancements have introduced alternative malaria detection methods, including rapid diagnostic tests (RDTs), polymerase chain reaction (PCR), and automated microscopy. RDTs and microscopic blood smear analysis remain widely used due to their reliability and practicality
[11]
. However, the choice of diagnostic tools depends on factors such as available healthcare infrastructure, personnel expertise, and malaria prevalence in a given area
[12]
.
Machine learning (ML), particularly Deep Learning (DL), has shown promise in automating malaria detection. Deep learning models, such as convolutional neural networks (CNNs), can analyze large datasets and identify malaria parasites in blood samples with high accuracy
[13]
. These techniques have also been used in population genetics to track disease patterns. Developing a malaria detection system based on deep learning can provide a more efficient, real-time, and accessible alternative to traditional methods. This approach can enhance early diagnosis and disease management, particularly in resource-limited regions.
Figure 2. Comparison of DL and traditional ML.
Figure 2 compares deep learning and traditional machine learning techniques for malaria parasite classification. The deep learning approach follows a simplified process where raw input data is directly fed into a deep learning model, which automatically learns features and patterns to produce an output result. This method eliminates the need for manual feature extraction, making it more efficient for complex image-based tasks such as malaria detection.
On the other hand, the traditional machine learning approach consists of multiple stages. It begins with data preprocessing to clean and standardize the input data, followed by manual feature extraction, where relevant characteristics of the data are identified. These extracted features are then used to establish a model, which is trained to classify malaria parasites. Finally, the model generates an output result.
The key advantage of deep learning over traditional machine learning is its ability to automatically learn features from raw data, reducing human intervention and improving accuracy in malaria parasite classification. However, deep learning models require large datasets and significant computational power for training. Traditional machine learning, while more interpretable and computationally less expensive, may struggle with complex feature learning.
Malaria continues to pose a serious threat to global health, particularly in low-income areas with weak healthcare systems
[5]
. Delayed diagnosis often leads to severe complications and increased mortality rates. While traditional diagnostic methods are effective, they can be slow, inaccessible, and prone to human error
[9]
. To overcome these limitations, this study proposes an automated malaria prediction model using deep learning. The goal is to develop an accurate and efficient system for detecting malaria parasites in blood samples, ensuring early diagnosis and improved treatment outcomes.
2. Related Works
Deep learning models have been developed to detect malaria parasites in blood samples, achieving an impressive accuracy of 99.68%. This approach outperforms previous methods in both accuracy and processing speed, making it a promising tool for malaria diagnosis, especially in regions with limited medical resources
[14]
.
VGG16 was utilized for malaria detection, delivering an accuracy of 97.60%, an AUC of 98.70%, an F1-score of 97.50%, a recall of 97.40%, and a precision of 97.60%. Although the model demonstrated solid performance, the impact of different image preprocessing techniques was not considered
[15]
.
Microscopic blood smear images were used to train an automated CNN-based malaria detection system. The model achieved an accuracy of 99.23% and was successfully deployed in both mobile and web-based environments, making it a valuable tool for low-resource settings
[16]
. To enhance detection performance, some studies incorporated ensemble learning techniques. A comparative study analyzing CNN, transfer learning, and Vision Transformers demonstrated notable improvements in malaria detection accuracy
[17]
.
Transfer learning methods have also played a crucial role in malaria detection. A pre-trained VGG model integrated with a Support Vector Machine significantly improved classification accuracy, showcasing the potential of transfer learning in medical image analysis
[18]
.
Accurately diagnosing mosquito-borne diseases remains a challenge due to symptoms that range from mild discomfort to severe complications. Traditional methods depend on experts analyzing blood smears, which is time-consuming and prone to errors. While machine learning has improved efficiency, it still struggles with complex pattern recognition. In contrast, deep learning automates feature extraction and enhances diagnostic precision. This study introduces EDRI, a hybrid deep learning model trained on the NIH Malaria dataset, achieving an accuracy of 97.68%, highlighting its potential as an efficient diagnostic tool for healthcare professionals
[19]
.
Mosquito-borne infections remain a serious health threat, necessitating more accurate and efficient diagnostic techniques. Traditional microscopy-based methods are prone to human error and require skilled technicians. This study presents three customized CNN models like PCNN, SPCNN, and SFPCNN to improve diagnostic accuracy. Among them, SPCNN performed best, achieving an accuracy of 99.37% and an AUC of 99.95%, with the fastest test time of 0.00252 seconds. Additionally, it outperformed existing transfer learning models such as VGG16 and ResNet152. Feature interpretation using Grad-CAM and SHAP demonstrated its ability to highlight infected regions, reinforcing the role of deep learning in disease detection
[20]
.
Identifying Plasmodium species accurately is crucial for effective treatment, yet traditional methods depend heavily on expert interpretation. CNN-based artificial intelligence models have significantly enhanced diagnostic accuracy by automating medical image analysis. However, distinguishing between P. falciparum and P. vivax remains a challenge. This study introduces a CNN model designed to classify infected and uninfected cells in thick blood smears. With a seven-channel input, the model achieved an accuracy of 99.51% and strong cross-validation performance. Future work will focus on adapting the system to real-world image quality to improve accessibility in remote areas
[21]
.
New diagnostic technologies offer alternatives to traditional methods like microscopy, PCR assays, and rapid tests, which require specialized skills. This study investigates mid-infrared (MIR) spectroscopy combined with machine learning to detect malaria using dried blood spots (DBS). Blood samples from 12 wards in southeastern Tanzania were analyzed using ATR-FTIR spectroscopy, producing high-resolution MIR spectra. After preprocessing, classification models were trained on PCR-confirmed data, with logistic regression proving most effective. It achieved 92% accuracy for P. falciparum and 85% for mixed infections, indicating MIR-ML's potential as a cost-effective, high-throughput screening tool, though further validation is needed
[22]
.
Timely malaria diagnosis remains difficult in resource-limited regions, where inadequate healthcare infrastructure and computing resources hinder rapid detection. Severe cases can be fatal within a week, emphasizing the need for accurate identification of parasite types and life cycle stages. This study introduces a deep learning model that classifies both parasite types and life cycle stages. The model is over 20 times lighter than DenseNet, with fewer than 0.4 million parameters, making it ideal for mobile applications. Tested on multiple public datasets, it demonstrated superior performance, supporting its potential deployment in malaria-endemic areas
[23]
.
Automated diagnostic systems significantly improve upon conventional microscopy, which is time-intensive and reliant on expert interpretation. This study presents an inception-based capsule neural network designed to enhance detection speed and accuracy. By integrating Inception V3 for feature extraction and a capsule network for classification, the model significantly outperformed traditional approaches. The results emphasize its potential for reliable malaria diagnosis in clinical settings
[24]
. Efficient malaria detection is essential, particularly in regions where it remains a leading cause of mortality. This study introduces a deep learning-based system for diagnosing malaria from peripheral blood smears. Comparing thin and thick smears, the results show that thick smears yield superior performance, achieving an accuracy of 96.97%. The proposed model surpasses established transfer learning approaches, proving effective for improving malaria diagnosis in endemic areas
[25]
.
Advancing deep learning-based diagnosis is vital for infectious disease management. This study develops a CNN model optimized for malaria detection using segmented image patches. Transfer learning with pre-trained CNN models such as VGG19, ResNet50, and MobileNetV2 improves classification accuracy. Evaluated on the NIH Malaria Dataset, the model achieved near-perfect accuracy of 100%, reinforcing its value as an advanced diagnostic tool
[26]
. Traditional diagnostic methods require expert analysis and can be error-prone, prompting the need for AI-driven solutions. This research explores convolutional neural networks (CNNs) for real-time malaria detection via mobile applications. Using a custom CNN model optimized with cyclical stochastic gradient descent (SGD), the system classifies infected and healthy red blood cells with 97.30% accuracy. The findings highlight deep learning’s potential in mobile-based malaria detection, particularly in areas with limited medical resources
[27]
.
The urgent need for rapid and accurate diagnostic tools has driven significant advancements in deep-learning models for detecting malaria. Traditional microscopic blood smear examination remains labor-intensive, necessitating automated solutions for improved efficiency. One study optimized the YOLOv4 model by introducing layer pruning and replacing the CSP-DarkNet53 backbone with ResNet50, achieving a mean average precision (mAP) of 90.70%. This refined model, YOLOv4-RC3_4, outperformed the original by over 9%, reducing computational complexity by 22% and model size by 23MB while enhancing infected cell detection by 9.27%
[28]
. Another research effort applied transfer learning and snapshot ensembling to classify malaria parasites in thin blood smear images. The snapshot ensembling model, leveraging the EfficientNet-B0 architecture, achieved exceptional performance, recording an F1 score of 99.37%, precision of 99.52%, and recall of 99.23%. GradCAM visualization further improved model transparency, highlighting parasitic regions and reinforcing trust in deep learning-based diagnostic applications
[29]
.
To address the limitations of manual screening, a Deep Boosted and Ensemble Learning (DBEL) framework was introduced. This system combined Boosted-BR-STM CNNs with ensemble machine learning classifiers, incorporating dilated-convolutional block-based Split Transform Merge (STM) and Squeezing–Boosting (SB) techniques. The framework excelled on the NIH malaria dataset, achieving 98.50% accuracy, 0.9920 sensitivity, and an AUC of 0.9960, demonstrating its potential for automated screening
[30]
. Comparing machine learning models, another study evaluated XG-Boost, support vector machines (SVM), and neural networks for malaria detection. SVM exhibited superior accuracy (94%) compared to XG-Boost (90%) and neural networks (80%), while CNNs emerged as the most effective, achieving 97% accuracy in identifying parasitized cells
[31]
.
Further innovations extended to malaria species classification using CNNs for feature extraction and SVM for categorization into four species: Plasmodium falciparum, P. vivax, P. ovale, and P. malariae. The model demonstrated high accuracy, integrating batch normalization and ReLU activation to enhance performance. By utilizing smartphone-based image acquisition, this approach offers a cost-effective diagnostic solution, particularly for resource-limited settings
[32]
.
While deep learning models have demonstrated high accuracy in malaria detection, real-time deployment remains a challenge, particularly in resource-limited environments. Many existing models, including VGG16, ResNet, and EfficientNet, achieve over 99% accuracy, but their large computational requirements and slow inference times hinder practical use in mobile and web applications. Additionally, variations in blood smear image quality and preprocessing inconsistencies impact model performance in real-world settings.
To overcome these challenges, AutoMalariaNet leverages VGG16 with optimized computational efficiency to ensure faster inference while maintaining high accuracy. The model will integrate lightweight techniques such as quantization, pruning, and knowledge distillation, reducing hardware demands without compromising performance. Additionally, an automated image preprocessing pipeline will standardize input quality, improving robustness across different clinical environments. Designed for seamless mobile and web integration, AutoMalariaNet will provide real-time, high-speed malaria detection, making it a practical and scalable solution for healthcare professionals in endemic regions.
3. Materials and Methods
This section outlines the materials and methodologies employed in conducting the experimental Machine Learning (ML) research. It details the experimental setup, data sources, preprocessing techniques, feature extraction methods, model selection, training process, evaluation metrics, and deployment strategies. The study follows a structured approach, incorporating dimensionality reduction, cross-validation, and hyperparameter tuning to enhance model performance. Finally, the optimized model is integrated into a security framework for real-time detection, ensuring a comprehensive and effective ML-based solution for identifying XSS attacks.
3.1. Experimental Setup
Table 1 presents the experimental setup used in this study. The system is powered by an Intel (R) Core™ i7-7300U processor, operating on a 64-bit Windows 10 platform with a x64-based architecture. It is equipped with 16GB of RAM and a 256GB SSD, ensuring efficient data processing and model training. The development environment is Jupyter Notebook, which provides an interactive and flexible platform for executing Python scripts. The implementation is carried out using Python 3.11, leveraging several essential libraries, including Pandas for data manipulation, NumPy for numerical computations, Scikit-learn for machine learning operations, TensorFlow for deep learning model development, Matplotlib for data visualization, and Seaborn for statistical plotting. This setup facilitates a robust and efficient framework for conducting machine learning experiments and implementing data-driven solutions.
Table 1. Experimental Setup.

Configuration

Parameters

CPU

Intel (R) Core (TM) i7-7300U

System Type

64-bit Operating System, x64-based processor

Memory (RAM)

16GB

Harddisk

256 GB SSD

Operating System (OS)

Microsoft Windows 10

Development IDE

Jupyter Notebook

Programming Language

Python 3.11

Package/Library

Pandas, Numpy, Scikit-learn, Tensorflow, Matplotlib, Seaborn
3.2. Proposed System Architecture
This section describes the architectural framework of the proposed system for malaria parasite detection, which leverages the VGG-16 model.
Figure 3. Proposed Deep Learning Pipeline for Malaria Parasite Detection.
3.2.1. Building Blocks / Training Parameters
Table 2. CNN Building Block.

Layer Type

Filter Size

Number of Filters

Activation

Other Parameters

Conv2D

(3,3)

32

ReLU

Input shape: (128, 128, 3)

MaxPooling2D

(2,2)

-

-

-

Conv2D

(3,3)

64

ReLU

-

MaxPooling2D

(2,2)

-

-

-

Conv2D

(3,3)

128

ReLU

-

MaxPooling2D

(2,2)

-

-

-

Flatten

-

-

-

Converts 2D feature maps into a 1D vector

Dense

-

256

ReLU

Fully connected layer

Dropout

-

-

-

Dropout rate: 0.5

Dense

-

1

Sigmoid

Output layer for binary classification
Table 3. CNN Model Training Parameters.

Parameter

Value

Optimizer

Adam

Learning Rate

0.0001

Loss Function

Binary Crossentropy

Metrics

Accuracy

Batch Size

32

Image Size

(128, 128)

Epochs

50

Data Augmentation

Yes (Rotation, Shift, Shear, Zoom, Flip)
Table 4. Proposed VGG-16 Model Building Block.

Layer Type

Number of Filters/ Units

Kernel Size

Activation Function

Other Parameters

VGG16 Base Model

Pretrained on ImageNet

-

-

Feature Extractor (Frozen)

Flatten

-

-

-

Converts features to 1D

Dense

256

-

ReLU

Fully Connected Layer

Dropout

-

-

-

50% Dropout

Dense

1

-

Sigmoid

Output Layer (Binary Classification)
Table 5. Proposed VGG-16 Model Training Parameters.

Parameter

Value

Optimizer

Adam

Learning Rate

0.0001

Loss Function

Binary Crossentropy

Metrics

Accuracy

Batch Size

32

Input Image Size

(128, 128, 3)

Epochs

50

Data Augmentation

Yes (Rotation, Width Shift, Height Shift, Shear, Zoom, Horizontal Flip)

Training Split

80%

Test Split

20%
Figure 4. Proposed VGG-16 Architectural Design for Malaria Parasite Detection.
Figure 4 illustrates the proposed VGG-16 model architecture for malaria parasite detection using blood smear images. The model leverages a pretrained VGG-16 base model on ImageNet, acting as a frozen feature extractor to capture low-level and high-level features from input images. The extracted features are then passed through a Flatten layer, converting them into a 1D vector for further processing. Next, a fully connected Dense layer with 256 units and a ReLU activation function is applied, allowing the model to learn complex patterns in the data. To reduce overfitting, a Dropout layer with a 50% dropout rate is introduced. Finally, the architecture concludes with an Output Dense layer containing a single unit with a Sigmoid activation function, enabling binary classification of malaria-infected and healthy blood smear images. The structured design of this model ensures efficient feature extraction, classification accuracy, and robustness for malaria detection in medical imaging applications.
3.2.2. Data Source and Description
This section discusses the source and description of dataset used in this study. Table 6 outlines the details of the malaria dataset used in this study. The dataset, obtained from the Kaggle repository, consists of 550 instances with categorical features and has a file size of approximately 6.34MB. Among these instances, 311 are labeled as parasitic, while 239 are uninfected, indicating an imbalanced class distribution. Since there are no missing values, preprocessing will focus on addressing this imbalance. To prevent the model from being biased toward the majority class, data augmentation techniques will be applied. These techniques include rotation (randomly rotating images within a small angle range), flipping (applying horizontal or vertical flips), and scaling and cropping (resizing images and selecting random cropped sections). These transformations will help balance the dataset, enhance model robustness, and improve classification accuracy.
Table 6. Dataset Source and Description.

Parameters

Description

Dataset Name

Malaria Dataset

Dataset File Size

6.34MB

Source

Kaggle Repository

Link to dataset

https://www.kaggle.com/datasets/meetnagadia/malaria-dataset

Feature Type

Categorical

Number of Instances

550

Missing Values

None

Parasitic

311

Uninfected

239

Comment

Imbalance dataset in terms of the class distribution. Augmentation technique will be used to balance the class distribution
3.2.3. Mathematical Expression
Dataset Representation
Let D be the dataset containing blood smear images and labels:
D={(Xi,yi)∣i=1,2,…,N}(1)
where:
Xi ∈ Rh×w×c represents an input image with height h, width w, and c color channels
yi ∈ {0,1} is the label, where 0 denotes "uninfected" and 1 denotes "infected"
N is the total number of samples
Preprocessing
Each image undergoes normalization and augmentation:
xi′=Xi- μσ(2)
where μ and σ are the mean and standard deviation of the dataset
Feature Extraction using VGG-16 Model
The VGG-16 model consists of convolutional layers fc, max-pooling layers fp, and fully connected layers ffc.
Convolutional Layer
Each convolutional layer applies a filter W(i) with a bias b(i) and activation function ReLU:
Z(l)=fc(X(l))=ReLU(W(l)*X(l-1)+b(l))(3)
Max-Pooling Layer
Pooling reduces the spatial dimensions:
 X(l)= fp(Z(l)) = max. Z(l)(4)
(m,n) ∈ k×k(5)
Where k×k is the pooling window
Fully Connected Layers & Classification
Flattened features Xflat are passed through fully connected layers:
h=ffc(Xflat)=ReLU(WfcXflat+bfc)(6)
A softmax function outputs class probabilities:
P(y=k|X)=exp⁡(hx)∑j=0i  exp⁡(hx)(7)
Loss Function
The model is trained using cross-entropy loss:
L=-∑i=1NyilogP(yi)+(1–yi)log(1–P(yi))(8)
Model Optimization
The weights are updated using gradient descent (Adam optimizer):
W(i)⃖W(i)-ղ∂L∂W(L)(9)
where η is the learning rate
Model Evaluation
The model's performance is evaluated on an unseen test set using standard evaluation metrics, including accuracy, precision, recall, F1-score, and ROC-AUC, as mathematically defined in Equations (9) to (13). These metrics provide a comprehensive assessment of the model’s effectiveness in distinguishing between malaria-infected and uninfected blood smear images. Accuracy measures the overall correctness of predictions, while precision indicates the proportion of correctly identified malaria cases. Recall assesses the model’s ability to detect all actual malaria-infected samples, and the F1-score balances precision and recall for a more reliable evaluation. Finally, the ROC-AUC score quantifies the model's capability to differentiate between infected and uninfected samples across varying threshold values.
Accuracy=TP+TNTP + FN + FP + TN(10)
Precision=TPTP + FP(11)
Recall=TPTP + FN(12)
F1-score=2xPr⁡ecision x RecallPr⁡ecision + Recall(13)
ROCAUC=∫01TPR dFPR(14)
Deployment & Inference
Given a new input image Xnew, the trained model predicts:
ŷ = argmaxP(y = k ∣ Xnew)(15)
where ŷ is the predicted class
This formulation mathematically expresses the pipeline from data gathering to deployment for malaria detection using VGG-16 model.
4. Results
This section provides a concise and well-structured summary of the experimental results, as depicted in Figures 5–14 and detailed in Table 7.
4.1. Dataset Representation & Visualization
This section presents the malaria parasite dataset in tabular form and visualizes sample images to provide insights into the data distribution. The dataset consists of labeled blood smear images categorized into infected (parasitized) and uninfected samples. Figures 5 - 8 shows the key dataset attributes, including the total number of images, class distribution, image resolution, and dataset split (training, and test sets). These graphical representation helps ensure a balanced dataset, which is crucial for model generalization. These visual representations aid in understanding data patterns and potential challenges, such as class imbalance or noise, which could impact the performance of the VGG-16 deep learning model. By representing the dataset graphically and visualizing key features, this section provides a comprehensive overview of the data, forming the foundation for effective training and evaluation of the malaria detection model.
Figure 5. Class Imbalance in Malaria Original Dataset Represented using Pie Chart.
Figure 5 illustrates the class distribution in the original malaria dataset using a pie chart. The dataset comprises two distinct categories: Parasitized Blood Smear Images and Uninfected Blood Smear Images. The red section of the chart, representing Parasitized Blood Smear Images, contains 311 instances, whereas the yellow section, representing Uninfected Blood Smear Images, consists of 239 instances. This uneven distribution highlights a class imbalance, which can significantly impact the performance of machine learning models by causing bias toward the majority class. To address this issue, various techniques such as data augmentation, oversampling, or synthetic data generation may be employed to achieve a more balanced dataset and enhance the model's generalizability.
Figure 6 presents a comparative analysis of the malaria dataset before and after data augmentation. The Initial Dataset (blue bars) consists of a total of 550 images, with 311 Parasitic and 239 Uninfected blood smear images. However, due to the class imbalance observed in Figure 4, data augmentation techniques were applied to balance the dataset. Following augmentation, the dataset expanded to 622 images, as shown by the red bars. Specifically, the Uninfected class, which initially had only 239 images, was increased to 311, ensuring an equal representation of both Parasitic and Uninfected images. This balancing process is crucial for improving the performance of deep learning models by preventing bias toward the majority class and enhancing generalization across different image variations.
Figure 6. Malaria Dataset Representations (Before and After Augmentation).
Figure 7. Sample Augmented Malaria Dataset.
Figure 8. Dataset Split Ratio.
Figure 7 illustrates a subset of the augmented malaria dataset, showcasing variations of a single blood smear image after applying augmentation techniques. These transformations, which may include rotation, flipping, zooming, and contrast adjustments, are essential for enhancing the diversity of the dataset. By artificially increasing the dataset size through augmentation, the model is exposed to a broader range of image variations, improving its ability to generalize to new, unseen data. This process mitigates overfitting and ensures that the deep learning model learns robust and meaningful features rather than memorizing specific patterns from the limited original dataset.
Figure 8 presents the dataset split ratio used for training and testing the malaria classification model. Following augmentation, the dataset expanded to 622 images, which were divided into training and testing subsets to ensure effective model learning and evaluation. The dataset was split into 80% for training (marked in blue) and 20% for testing (marked in red). The training set, which consists of 497 images, was used to train the deep learning model by allowing it to learn relevant features from the blood smear images. The testing set, comprising 125 images, was reserved for evaluating the model’s performance on unseen data. This 80-20 split is a common practice in machine learning to balance sufficient training data while keeping a meaningful portion for model evaluation. It ensures that the model can generalize well to new images and is not overfitted to the training data.
4.2. VGG-16 Training Results
Figure 9 illustrates the training and validation accuracy trends of the VGG-16 model over 50 epochs. The blue line represents the training accuracy, while the orange line corresponds to the validation accuracy. Initially, the training accuracy starts at a lower value but improves progressively as the model learns patterns in the dataset. Around the 10th epoch, the accuracy stabilizes, oscillating between 0.74 and 0.78. The validation accuracy follows a similar trend but exhibits more fluctuations, suggesting some level of variance in performance on unseen data. The fluctuations in validation accuracy indicate that the model might be experiencing slight overfitting, where it performs well on training data but struggles to maintain consistency on validation data. However, the overall accuracy trend suggests that the model is learning effectively. Additional fine-tuning, such as regularization or dropout layers, could be employed to enhance generalization and minimize fluctuations.
Figure 9. Training Accuracy for VGG-16 Model.
Figure 10 presents the training and validation loss curves for the VGG-16 model over 50 epochs. The blue line represents the training loss, while the orange line corresponds to the validation loss. At the beginning of training, the model exhibits a high loss value, which rapidly decreases within the first few epochs as it learns meaningful patterns. The training loss continues to decline steadily, indicating improved error minimization on the training set. However, the validation loss shows significant fluctuations, suggesting potential overfitting. While the validation loss does not exhibit a consistently increasing pattern, its instability compared to training loss indicates inconsistent performance on unseen data. This discrepancy suggests the need for additional regularization techniques, such as dropout or weight decay, to enhance generalization. The model has effectively learned features from the dataset, but further fine-tuning is required to improve stability and mitigate overfitting.
Figure 10. Training Loss for VGG-16 Model.
4.3. Deep Learning Models Classification Result
Table 7 presents a summary of deep learning model classification results for malaria parasite detection using blood smear images. The VGG-16 model outperforms the standard CNN, achieving an accuracy of 97% compared to CNN’s 87%, demonstrating its superior ability to differentiate between infected and uninfected cells. Additionally, VGG-16 exhibits higher precision (96%), recall (96.56%), and F1-score (97%), indicating its effectiveness in minimizing false positives and false negatives while maintaining balanced performance. These results highlight VGG-16’s robustness in malaria detection, making it a more reliable choice for automated diagnosis. The significant performance gap suggests that advanced architectures with deeper layers and pre-trained features, such as VGG-16, can extract more intricate patterns in blood smear images, enhancing diagnostic accuracy. This improved classification capability is crucial for early malaria detection, aiding healthcare professionals in timely and precise treatment, particularly in resource-limited settings.
Table 7. Summary of DL Models Classification Results.

Model

Accuracy

Precision

Recall

F1-Score

CNN

87

86

85

84.45

VGG-16

97

96

96.56

97
4.4. Confusion Matrix / ROC_AUC Result for VGG-16 Model
Figure 11 presents the confusion matrix heatmap for the VGG-16 model applied to the malaria parasite dataset, highlighting its classification performance in distinguishing infected and uninfected blood smear images. The model correctly identified 64 uninfected samples (true negatives) and 56 infected samples (true positives), indicating strong classification capability. With only 2 misclassified uninfected images (false positives) and 2 misclassified infected images (false negatives), the model demonstrates high accuracy in detecting malaria parasites. The minimal misclassification suggests that VGG-16 effectively learns relevant features from the dataset, making it a reliable tool for malaria diagnosis. However, slight performance improvements through techniques such as data augmentation or regularization could further enhance the model's robustness.
Figure 11. Confusion Matrix Heatmap for VGG-16 Model.
Figure 12 illustrates the ROC curve for the VGG-16 model in classifying malaria parasite-infected and uninfected blood smear images, with an Area Under the Curve (AUC) of 0.97. This high AUC value indicates that the model has excellent discriminatory power, effectively distinguishing between positive (infected) and negative (uninfected) cases. The curve remains close to the top-left corner, signifying a low false positive rate and a high true positive rate, which are essential for malaria diagnosis to minimize misclassifications. Such strong performance suggests that VGG-16 is well-suited for malaria detection tasks, though further optimization, such as fine-tuning hyperparameters or employing additional feature extraction techniques, could further enhance its effectiveness in real-world applications.
Figure 12. ROC_AUC for VGG-16 Model.
4.5. Comparison Evaluation
Figure 13. Graphical Comparison Evaluation of DL Models.
Figure 13 presents a comparative evaluation of deep learning models for malaria parasite detection using key performance metrics. The results indicate that the VGG-16 model outperforms the conventional CNN across all metrics, achieving an accuracy of 97% compared to 87% for CNN. Similarly, VGG-16 records higher precision (96%), recall (96.56%), and F1-score (97%), while the CNN attains 86%, 85%, and 84.45%, respectively. The superior performance of VGG-16 suggests its enhanced feature extraction capability, enabling more accurate differentiation between malaria-infected and uninfected blood smear images. This highlights the effectiveness of transfer learning in improving malaria detection accuracy, making VGG-16 a more reliable model for automated malaria diagnosis in clinical settings.
4.6. Developed Web Interface
Figure 14. Web Result Interface for Malaria Detection.
Figure 14 showcases the web-based result interface for the VGG-16 deep-learning model in malaria parasite detection from blood smear images. The interface provides an automated diagnostic result, classifying the uploaded blood smear image as either infected or uninfected. By leveraging VGG-16’s high-performance feature extraction capabilities, the model ensures accurate and reliable malaria detection, aiding in rapid clinical decision-making. The user-friendly interface allows seamless image uploads and immediate classification feedback, making it suitable for real-world applications in healthcare settings. Such an automated system enhances early malaria diagnosis, potentially reducing misdiagnosis and improving treatment outcomes, particularly in resource-limited regions where expert pathologists may not always be available.
5. Discussion
This section provides a comparative analysis and discussion of the research findings and previous studies in terms of dataset characteristics, classification performance, model effectiveness, and emerging trends in malaria parasite detection.
5.1. Model Performance Comparison
The findings from this study highlight the effectiveness of the VGG-16 model in malaria parasite detection using blood smear images, achieving an accuracy of 97%, an F1-score of 97%, and an AUC of 0.97.
When compared to previous studies, several key observations emerge. The study by
[14]
reported a model with an accuracy of 99.68%, surpassing the 97% accuracy of VGG-16 in this study. However, details on dataset balancing and preprocessing techniques were not explicitly discussed, which could have influenced the results. Similarly, the work in
[15]
demonstrated a VGG-16-based approach achieving slightly higher accuracy (97.60%) and AUC (98.70%). This suggests that alternative preprocessing techniques could further improve the performance of VGG-16 in malaria detection.
The CNN-based model in
[16]
achieved an accuracy of 99.23% and was successfully deployed in mobile and web-based applications, emphasizing the significance of accessibility in malaria diagnosis. Furthermore, studies integrating ensemble learning
[17]
and hybrid deep learning models such as EDRI
[19]
attained even higher performance, indicating that leveraging multiple architectures can enhance malaria detection accuracy.
5.2. Dataset Characteristics and Diversity
The dataset used in this study consists of blood smear images, with class balancing achieved through augmentation techniques. The initial dataset contained 550 images, which were increased to 622 through augmentation to address class imbalance.
In comparison, the dataset used in
[15]
was significantly larger, with over 27,000 images, potentially contributing to the slightly higher reported accuracy. However, the impact of dataset diversity on generalization was not explicitly analyzed in that study. Meanwhile,
[17]
utilized a publicly available dataset combined with a locally collected dataset, demonstrating that dataset diversity can enhance model robustness and performance.
These observations highlight the importance of dataset size and diversity in training deep learning models, emphasizing the need for larger and more varied datasets to improve generalization across different populations and imaging conditions.
5.3. Preprocessing and Augmentation
This study addressed class imbalance through data augmentation, increasing the dataset size from 550 to 622 images. This augmentation strategy contributed to improved model generalizability and robustness.
In contrast, related works such as
[15]
did not explicitly analyze the effect of preprocessing on model performance. The absence of such an analysis could be a factor influencing the discrepancies in accuracy across studies.
Additionally, the introduction of feature augmentation (Figure 7) in this study contributed to robust learning, mitigating overfitting. This challenge was observed in
[18]
, where transfer learning alone was insufficient to achieve optimal classification. These results suggest that preprocessing strategies play a crucial role in determining the effectiveness of deep learning models for malaria detection.
5.4. Model Effectiveness and Generalization
The results indicate that VGG-16 exhibits strong feature extraction capabilities, outperforming traditional CNN models (97% vs. 87%). This finding aligns with previous studies that have shown the superiority of transfer learning-based architectures over conventional CNN models in medical image classification. Moreover, studies incorporating hybrid models, such as
[19]
, achieved slightly better accuracy (97.68%). This suggests that combining deep learning with additional feature engineering or hybrid architectures could further optimize malaria detection.
Additionally, the confusion matrix (Figure 11) from this study revealed a high true positive rate, indicating that VGG-16 performs well in distinguishing between parasitized and non-parasitized cells. This level of generalization supports its potential for real-world deployment in malaria screening applications.
5.5. Way Forward
Building on the comparative analysis, future research should focus on developing hybrid deep learning models that integrate feature engineering and ensemble learning techniques to further enhance malaria detection accuracy and robustness. By combining multiple architectures, such as convolutional neural networks (CNNs) with Vision Transformers (ViTs) or attention mechanisms, models can capture both spatial and contextual features more effectively, leading to improved generalization.
Moreover, incorporating Explainable AI (XAI) techniques will be crucial in making the system more transparent and interpretable, increasing its trustworthiness for clinical applications. Methods such as Gradient-weighted Class Activation Mapping (Grad-CAM) can provide heatmap visualizations that highlight the specific regions in an image responsible for model predictions. This will aid medical professionals in verifying the reliability of AI-generated diagnoses and ensuring that the system aligns with medical decision-making processes.
Additionally, optimizing data preprocessing strategies such as advanced augmentation techniques, noise reduction, and adaptive contrast enhancement can further refine model performance by ensuring that input images are more representative of real-world conditions. Future research should also explore alternative deep learning architectures, including Large Language Models (LLMs) that can integrate textual clinical notes with image analysis, as well as ensemble deep learning approaches that combine multiple model predictions to reduce bias and variance.
Lastly, to ensure the system's applicability in diverse real-world settings, future studies should extend evaluations to larger and more diverse datasets, covering variations in image quality, patient demographics, and malaria strains. This will help assess the model’s robustness across different populations and imaging conditions, ensuring its effectiveness in clinical deployments. By addressing these aspects, the proposed system can evolve into a more reliable, interpretable, and scalable diagnostic tool for malaria detection in both low-resource and advanced healthcare settings.
6. Conclusions
This study demonstrated the effectiveness of the VGG16 deep learning model for malaria detection using blood smear images, achieving a high classification accuracy of 97%. The results underscore the model’s capability in extracting significant features, making it a reliable tool for automated malaria diagnosis. Compared to existing methods, the proposed model performed competitively, reinforcing the potential of deep learning in medical image analysis. Additionally, the study highlighted the critical role of preprocessing techniques, such as data augmentation, in improving model generalizability and addressing class imbalance, ensuring more robust performance across varying data distributions.
While the results are promising, further advancements are necessary to enhance classification accuracy and model interpretability. Future research should explore hybrid deep learning models that integrate feature engineering and ensemble learning techniques to improve performance. Investigating alternative architectures, such as Vision Transformers (ViTs) and Large Language Models (LLMs), could optimize malaria detection by capturing richer spatial and contextual information. Moreover, integrating Explainable AI (XAI) techniques like Class Activation Mapping (CAM) and Grad-CAM will improve transparency, allowing medical professionals to understand the reasoning behind model predictions and facilitating clinical adoption.
These contributions add to the growing body of knowledge on deep learning-based malaria detection, offering valuable insights into model effectiveness and generalization. Extending this research to larger and more diverse datasets will be essential for evaluating the model’s robustness across different populations and imaging conditions, ensuring its applicability in real-world clinical settings. By incorporating these advancements, future malaria detection systems can become more accurate, interpretable, and scalable, ultimately enhancing diagnostic efficiency and accessibility in resource-constrained healthcare environments.
Abbreviations

ML

Machine Learning

AI

Artificial Intelligence

DL

Deep Learning

ANN

Artificial Neural Network

CNN

Convolutional Neural Networks

TL

Transfer Learning

EL

Ensemble Learning

CDC

Center for Disease Control

RDT

Rapid Diagnostic Test

PCR

Polymerase Chain Reaction

DBS

Dried Blood Spot

mAP

Mean Average Precision

DBEL

Deep Boosted and Ensemble Learning

SVM

Support Vector Machine

YOLOv4

You Only Look Once version 4

ATR-FTIR

Attenuated Total Reflectance - Fourier Transform Infrared Spectroscopy

MIR

Mid-infrared Spectrum in spectroscopy

SGD

Stochastic Gradient Descent

SHAP

SHapley Additive exPlanations

AUC

Area Under the Curve

SFPCNN

Sparse Fully Parallel Convolutional Neural Network

SPCNN

Spectral-Partition Convolutional Neural Network

PCNN

Pulse-Coupled Neural Network

EDRI

Economic Development and Research Institute

VGG-16

Visual Geometry Group 16-layer Model
Author Contributions
Emmanuel Osaze Oshoiribhor: Conceptualization, Resources, Methodology, Formal Analysis, Validation, original draft, review & editing
Adetokunbo MacGregor John-Otumu: Data curation, Methodology, Formal Analysis, Validation, Visualization, review & editing
Data Availability Statement
The data that support the findings of this study can be found at: https://www.kaggle.com/datasets/meetnagadia/malaria-dataset (a publicly available repository url)
Conflicts of Interest
The authors declare no conflicts of interest.
References
[1] González-Sanz, M., Berzosa, P., & Norman, F. F. (2023). Updates on Malaria Epidemiology and Prevention Strategies. Current infectious disease reports, 1–9. Advance online publication.

https://doi.org/10.1007/s11908-023-00805-9

[2] Al-Awadhi, M., Ahmad, S., & Iqbal, J. (2021). Current Status and the Epidemiology of Malaria in the Middle East Region and Beyond. Microorganisms, 9(2), 338.

https://doi.org/10.3390/microorganisms9020338

[3] Mawson, A. R. (2013). The pathogenesis of malaria: A new perspective. Pathogens and Global Health, 107(3), 122–129.

https://doi.org/10.1179/2047773213Y.0000000084

[4] Sato, S. (2021). Plasmodium falciparum and severe malaria: Mechanisms and implications. Malaria Research Journal, 19(3), 321-335.
[5] Oshagbemi, O. A., Lopez-Romero, P., Winnips, C., & others. (2023). Estimated distribution of malaria cases among children in sub-Saharan Africa by specified age categories using data from the Global Burden of Diseases 2019. Malaria Journal, 22, 371.

https://doi.org/10.1186/s12936-023-04811-z

[6] World Health Organization. (2020). Malaria diagnosis, treatment, and preventive therapies. Retrieved from

https://www.who.int/malaria/publications

[7] Ahuru, R. (2018). Economic impact of malaria on households in Nigeria. African Economic Review, 12(4), 312-327.
[8] Alonso, S., Chaccour, C. J., Elobolobo, E., & others. (2019). The economic burden of malaria on households and the health system in a high transmission district of Mozambique. Malaria Journal, 18, 360.

https://doi.org/10.1186/s12936-019-2995-4

[9] Yalley, A. K., Ocran, J., Cobbinah, J. E., Obodai, E., Yankson, I. K., Kafintu-Kwashie, A. A., Amegatcher, G., Anim-Baidoo, I., Nii-Trebi, N. I., & Prah, D. A. (2024). Advances in malaria diagnostic methods in resource-limited settings: A systematic review. Tropical Medicine and Infectious Disease, 9(9), 190.

https://doi.org/10.3390/tropicalmed9090190

[10] Al Kuwaiti, A., Nazer, K., Al-Reedy, A., Al-Shehri, S., Al-Muhanna, A., Subbarayalu, A. V., Al Muhanna, D., & Al-Muhanna, F. A. (2023). A Review of the Role of Artificial Intelligence in Healthcare. Journal of Personalized Medicine, 13(6), 951.

https://doi.org/10.3390/jpm13060951

[11] Syeda, S. M., Rania, M., Rehab, S., Mirza, A. B., & Swamy, K. (2022). Comparative analysis of the diagnostic accuracy of malaria parasite by microscopy, RDTs (PfHRP-2 & PLDH), and PCR. Indian Journal of Pathology and Oncology, 9(2), 112–115.

https://doi.org/10.18231/j.ijpo.2022.028

[12] Mosquera Romero, M., Zuluaga Idárraga, L., & Tobón Castaño, A. (2018). Malaria Journal, 17(440).

https://doi.org/10.1186/s12936-018-2591-z

[13] Jdey, I., Hcini, H., & Ltifi, H. (2023). Deep learning and machine learning for malaria detection: Overview, challenges, and future directions. International Journal of Information Technology & Decision Making, 23(5), 1–32.

https://doi.org/10.1142/S0219622023300045

[14] Siłka, W., Wieczorek, M., Siłka, J., & Woźniak, M. (2023). Malaria detection using advanced deep learning architecture. Sensors, 23(3), 1501.

https://doi.org/10.3390/s23031501

[15] Yang, F., Poostchi, M., Yu, H., Zhou, Z., Silamut, K., Yu, J., Maude, R. J., Jaeger, S., & Antani, S. (2020). Deep Learning for Smartphone-Based Malaria Parasite Detection in Thick Blood Smears. IEEE journal of biomedical and health informatics, 24(5), 1427–1438.

https://doi.org/10.1109/JBHI.2019.2939121

[16] Fuhad, K. M. F., Tuba, J. F., Sarker, M. R. A., Momen, S., Mohammed, N., & Rahman, T. (2020). Deep Learning Based Automatic Malaria Parasite Detection from Blood Smear and its Smartphone Based Application. Diagnostics (Basel, Switzerland), 10(5), 329.

https://doi.org/10.3390/diagnostics10050329

[17] Poojary, H. D., & Sumithra, T. V. (2022). Comparative Analysis of Deep Learning Models for Malaria Detection. 1–6.

https://doi.org/10.1109/GCAT55367.2022.9972167

[18] Vijayalakshmi, A., & Rajesh Kanna, B. (2019). Deep learning approach to detect malaria from microscopic images. Multimedia Tools and Applications, 79, 15297–15317.

https://doi.org/10.1007/s11042-019-7162-y

[19] Boit, S., & Patil, R. (2024). An efficient deep learning approach for malaria parasite detection in microscopic images. Diagnostics, 14(23), 2738.

https://doi.org/10.3390/diagnostics14232738

[20] Ahamed, M. F., Nahiduzzaman, M., Mahmud, G., Shafi, F. B., Ayari, M. A., Khandakar, A., Abdullah-Al-Wadud, M., & Islam, S. M. R. (2025). Improving malaria diagnosis through interpretable customized CNN architectures. Scientific Reports, 15(1), 6484.

https://doi.org/10.1038/s41598-025-90851-1

[21] Ramos-Briceño, D. A., Flammia-D’Aleo, A., Fernández-López, G., et al. (2025). Deep learning-based malaria parasite detection: Convolutional neural networks model for accurate species identification of Plasmodium falciparum and Plasmodium vivax. Scientific Reports, 15, 3746.

https://doi.org/10.1038/s41598-025-87979-5

[22] Mwanga, E. P., Minja, E. G., Mrimi, E., et al. (2019). Detection of malaria parasites in dried human blood spots using mid-infrared spectroscopy and logistic regression analysis. Malaria Journal, 18, 341.

https://doi.org/10.1186/s12936-019-2982-9

[23] Chaudhry, H. A. H., Farid, M. S., Fiandrotti, A., et al. (2024). A lightweight deep learning architecture for malaria parasite-type classification and life cycle stage detection. Neural Computing & Applications, 36, 19795–19805.

https://doi.org/10.1007/s00521-024-10219-w

[24] Madhu, G., Mohamed, A. W., Kautish, S., et al. (2023). Intelligent diagnostic model for malaria parasite detection and classification using imperative inception-based capsule neural networks. Scientific Reports, 13, 13377.

https://doi.org/10.1038/s41598-023-40317-z

[25] Uzun Ozsahin, D., Mustapha, M. T., Bartholomew Duwa, B., & Ozsahin, I. (2022). Evaluating the performance of deep learning frameworks for malaria parasite detection using microscopic images of peripheral blood smears. Diagnostics, 12, 2702.

https://doi.org/10.3390/diagnostics12112702

[26] Alnussairi, M. H. D., & İbrahim, A. A. (2022). Malaria parasite detection using deep learning algorithms based on CNNs technique. Computers and Electrical Engineering, 103, 108316.

https://doi.org/10.1016/j.compeleceng.2022.108316

[27] Masud, M., Alhumyani, H. A., Alshamrani, S. S., Cheikhrouhou, O., Ibrahim, S., Ibrahim, S., Muhammad, G., Hossain, M. S., & Shorfuzzaman, M. (2020). Leveraging deep learning techniques for malaria parasite detection using mobile application. Wireless Communications and Mobile Computing, 2020, 8895429: 1–8895429: 15.

https://doi.org/10.1155/2020/8895429

[28] Sukumarran, D., Hasikin, K., Khairuddin, A. S. M., et al. (2024). An optimised YOLOv4 deep learning model for efficient malarial cell detection in thin blood smear images. Parasites & Vectors, 17, 188.

https://doi.org/10.1186/s13071-024-06215-7

[29] Mishra, S. (2021). Malaria parasite detection using efficient neural ensembles. Journal of Electronics, Electromedical Engineering, and Medical Informatics, 3(3), 119-133.

https://doi.org/10.35882/jeeemi.v3i3.2

[30] Asif, H. M., Khan, S. H., Alahmadi, T. J., et al. (2024). Malaria parasitic detection using a new deep boosted and ensemble learning framework. Complex & Intelligent Systems, 10, 4835–4851.

https://doi.org/10.1007/s40747-024-01406-2

[31] Alharbi, A. H., Aravinda, C. V., Lin, M., Ashwini, B., Jabarulla, M. Y., & Shah, M. A. (2022). Detection of peripheral malarial parasites in blood smears using deep learning models. Computational Intelligence and Neuroscience, 2022, 1-7.

https://doi.org/10.1155/2022/3922763

[32] Shashikiran, S., Babu, N. S., Rama, P., Mondal, S., Sadaqathulla, S., & Pramod, K. S. (2024). Malaria parasite detection using advanced deep learning techniques. In 2024 International Conference on Knowledge Engineering and Communication Systems (ICKECS), Chikkaballapur, India (pp. 1-4).

https://doi.org/10.1109/ICKECS61492.2024.10616908

Cite This Article
Plain Text BibTeX RIS

  • APA Style

    Oshoiribhor, E. O., John-Otumu, A. M. (2025). AutoMalariaNet: A VGG16-Based Deep Learning Model for High-Performance Automated Malaria Parasite Detection in Blood Smear Images. American Journal of Neural Networks and Applications, 11(1), 11-27. https://doi.org/10.11648/j.ajnna.20251101.12

    Copy | Download

    ACS Style

    Oshoiribhor, E. O.; John-Otumu, A. M. AutoMalariaNet: A VGG16-Based Deep Learning Model for High-Performance Automated Malaria Parasite Detection in Blood Smear Images. Am. J. Neural Netw. Appl. 2025, 11(1), 11-27. doi: 10.11648/j.ajnna.20251101.12

    Copy | Download

    AMA Style

    Oshoiribhor EO, John-Otumu AM. AutoMalariaNet: A VGG16-Based Deep Learning Model for High-Performance Automated Malaria Parasite Detection in Blood Smear Images. Am J Neural Netw Appl. 2025;11(1):11-27. doi: 10.11648/j.ajnna.20251101.12

    Copy | Download 

  • @article{10.11648/j.ajnna.20251101.12,
  author = {Emmanuel Osaze Oshoiribhor and Adetokunbo MacGregor John-Otumu},
  title = {AutoMalariaNet: A VGG16-Based Deep Learning Model for High-Performance Automated Malaria Parasite Detection in Blood Smear Images
},
  journal = {American Journal of Neural Networks and Applications},
  volume = {11},
  number = {1},
  pages = {11-27},
  doi = {10.11648/j.ajnna.20251101.12},
  url = {https://doi.org/10.11648/j.ajnna.20251101.12},
  eprint = {https://article.sciencepublishinggroup.com/pdf/10.11648.j.ajnna.20251101.12},
  abstract = {This research paper presents an automated malaria detection system using deep learning techniques to enhance diagnostic accuracy and efficiency, addressing the critical challenge of early and precise malaria diagnosis, especially in resource-constrained regions. Malaria remains a significant global health burden, particularly in tropical and subtropical regions where timely and accurate diagnosis is crucial for effective treatment and control. Traditional diagnostic methods, such as microscopic examination of blood smears, require skilled parasitologists and are often labor-intensive and time-consuming, making rapid detection difficult. To overcome these limitations, this study develops a deep learning-based malaria detection system integrating a Custom Convolutional Neural Network (CNN) and a pre-trained VGG16 model, trained on a publicly available malaria blood smear image dataset from Kaggle. Several data preprocessing techniques, including normalization and augmentation (rotation, flipping, scaling, and brightness adjustment), were applied to improve model generalization and robustness. The system is deployed through a web-based interface developed using Python, Flask, and HTML, allowing users to upload blood smear images and obtain real-time diagnostic results. Experimental evaluations demonstrate that the VGG16 model outperforms the Custom CNN, achieving an accuracy of 97%, precision of 96%, recall of 96.56%, and an F1-score of 97%, whereas the Custom CNN attained an accuracy of 87%, precision of 86%, recall of 85%, and an F1-score of 84.45%. These findings validate the effectiveness of deep learning in automating malaria detection and reducing reliance on manual microscopic examination, offering a scalable and accessible diagnostic tool for healthcare facilities with limited resources. Despite the success of the proposed system, further research is necessary to enhance model interpretability and trustworthiness. Future work should explore the integration of Vision Transformers (ViTs), Large Language Models (LLMs), and Ensemble Deep Learning techniques to improve malaria detection performance. Additionally, Explainable AI (XAI) methods, such as Grad-CAM, should be incorporated to provide visual explanations of model predictions, ensuring transparency and aiding medical professionals in understanding the decision-making process. By integrating these advancements, future systems can enhance both diagnostic accuracy and interpretability, making AI-driven malaria detection more reliable and widely applicable.
},
 year = {2025}
}

    Copy | Download 

  • TY  - JOUR
T1  - AutoMalariaNet: A VGG16-Based Deep Learning Model for High-Performance Automated Malaria Parasite Detection in Blood Smear Images

AU  - Emmanuel Osaze Oshoiribhor
AU  - Adetokunbo MacGregor John-Otumu
Y1  - 2025/04/17
PY  - 2025
N1  - https://doi.org/10.11648/j.ajnna.20251101.12
DO  - 10.11648/j.ajnna.20251101.12
T2  - American Journal of Neural Networks and Applications
JF  - American Journal of Neural Networks and Applications
JO  - American Journal of Neural Networks and Applications
SP  - 11
EP  - 27
PB  - Science Publishing Group
SN  - 2469-7419
UR  - https://doi.org/10.11648/j.ajnna.20251101.12
AB  - This research paper presents an automated malaria detection system using deep learning techniques to enhance diagnostic accuracy and efficiency, addressing the critical challenge of early and precise malaria diagnosis, especially in resource-constrained regions. Malaria remains a significant global health burden, particularly in tropical and subtropical regions where timely and accurate diagnosis is crucial for effective treatment and control. Traditional diagnostic methods, such as microscopic examination of blood smears, require skilled parasitologists and are often labor-intensive and time-consuming, making rapid detection difficult. To overcome these limitations, this study develops a deep learning-based malaria detection system integrating a Custom Convolutional Neural Network (CNN) and a pre-trained VGG16 model, trained on a publicly available malaria blood smear image dataset from Kaggle. Several data preprocessing techniques, including normalization and augmentation (rotation, flipping, scaling, and brightness adjustment), were applied to improve model generalization and robustness. The system is deployed through a web-based interface developed using Python, Flask, and HTML, allowing users to upload blood smear images and obtain real-time diagnostic results. Experimental evaluations demonstrate that the VGG16 model outperforms the Custom CNN, achieving an accuracy of 97%, precision of 96%, recall of 96.56%, and an F1-score of 97%, whereas the Custom CNN attained an accuracy of 87%, precision of 86%, recall of 85%, and an F1-score of 84.45%. These findings validate the effectiveness of deep learning in automating malaria detection and reducing reliance on manual microscopic examination, offering a scalable and accessible diagnostic tool for healthcare facilities with limited resources. Despite the success of the proposed system, further research is necessary to enhance model interpretability and trustworthiness. Future work should explore the integration of Vision Transformers (ViTs), Large Language Models (LLMs), and Ensemble Deep Learning techniques to improve malaria detection performance. Additionally, Explainable AI (XAI) methods, such as Grad-CAM, should be incorporated to provide visual explanations of model predictions, ensuring transparency and aiding medical professionals in understanding the decision-making process. By integrating these advancements, future systems can enhance both diagnostic accuracy and interpretability, making AI-driven malaria detection more reliable and widely applicable.

VL  - 11
IS  - 1
ER  -

    Copy | Download

Author Information

  • Emmanuel Osaze Oshoiribhor

    Department of Computer Science, Ambrose Alli University, Ekpoma, Nigeria

    Biography: Emmanuel Osaze Oshoiribhor is a Senior Lecturer in the Department of Computer Science at Ambrose Alli University. He earned his PhD in Computer Science from Ambrose Alli University in 2017, following his Master of Science and Bachelor of Science degrees from the University of Benin. As an active member of several professional organizations, including the Nigeria Computer Society (NCS), he has con-tributed to the academic community through participation in multiple international conferences. His research interests focus on Investigative Data Mining, Data Science, and Machine Learning.

    Research Fields: Investigative Data Mining, Machine Learning, Data Science.

    Contact Email

    http://orcid.org/0009-0004-8777-2050

  • Adetokunbo MacGregor John-Otumu

    Department of Information Technology, Federal University of Technology, Owerri, Nigeria

    Biography: Adetokunbo MacGregor John-Otumu is a Lecturer and Researcher in the Department of Information Technology at the Federal Univer-sity of Technology Owerri (FUTO). He was a Postdoctoral Research Fellow at Morgan State University, USA, from 2021 to 2022. He earned a Ph.D. in Computer Science (Artificial Intelligence) from Ebonyi State University in 2018, along with master’s degrees in Computer Science (Ambrose Alli University) and Information Technology (National Open University of Nigeria). Dr. John-Otumu is a member of professional bodies such as NCS, CPN, IEEE, ACM, AAAI, and the Internet Society. He leads the Machine Learning Research Group in his Department, at FUTO and contributes to global research collaborations. In 2024, he was a Keynote Speaker at the Climate Change Summit (NCS, Imo State) and served as a Technical Committee Member and Session Chair at FUTO’s first International Conference on ICT. He is also an editorial board member for several academic publications.

    Research Fields: Computer Vision, Deep Learning, Machine Learning, NLP, Health Informatics.

    Contact Email

    http://orcid.org/0000-0002-3138-4639

Download PDF
Submit an Article
  • Abstract
  • Keywords

  • Document Sections

    1. 1. Introduction
    2. 2. Related Works
    3. 3. Materials and Methods
    4. 4. Results
    5. 5. Discussion
    6. 6. Conclusions
    Show Full Outline
  • Abbreviations
  • Author Contributions
  • Data Availability Statement
  • Conflicts of Interest
  • References
  • Cite This Article
  • Author Information

About Us

Science Publishing Group (SciencePG) is an Open Access publisher, with more than 300 online, peer-reviewed journals covering a wide range of academic disciplines.

Learn More About SciencePG

Products

Information

Important Link

Copyright © 2012 -- 2026 Science Publishing Group – All rights reserved.