Deep Learning Model for Improved Malaria Prediction and Severity Classification

Abstract

Malaria remains a major health problem worldwide, especially in poor countries where quick and accurate diagnosis can save lives. This study presents a deep learning approach to improve how we detect malaria and measure how serious the infection is. We used a type of artificial intelligence called Convolutional Neural Networks (CNNs) to look at images of blood cells taken from microscopes and identify malaria parasites in them. Our model can not only tell if a cell is infected or not; it can also figure out how severe the infection is by counting how many parasites are present. To make the model work better, we applied a preprocessing technique called Gaussian blur, which helps reduce noise in the images and makes important features clearer. We tested the model using different training-to-testing splits (90:10, 80:20, 70:30, 60:40, and 50:50) and compared its performance with and without Gaussian blur. The CNN model with Gaussian blur achieved the best accuracy of 94.97% on the 90:10 data split. When we compared the CNN with other common machine learning models; Random Forest, Decision Tree, and Logistic Regression; the CNN performed far better, showing that deep learning is a powerful tool for malaria diagnosis. The model also successfully classified infection severity as mild, moderate, or severe based on parasitemia levels (the percentage of infected red blood cells). These results suggest that CNN-based systems could provide fast, reliable, and affordable malaria diagnosis in places where expert microscopists are scarce. Future work should focus on testing the model with larger and more diverse datasets and deploying it on mobile or web platforms for real-world use.