Evaluation of the Performance of Deep Learning Classification Models in Microscopic Image Results for Malaria Disease

Malaria is a significant public health issue, particularly in Sub-Saharan Africa, and is caused by the Plasmodium parasite, which is transmitted through mosquito bites. Traditionally, malaria diagnosis relies on the microscopic analysis of blood smears, a method that's dependable but also time-consuming and demanding in expertise. This study introduces and assesses a deep learning classification model using VGG19, aimed at automating malaria diagnosis in microscopic images. The model generated encouraging results, with a training accuracy of 97.76%, a validation accuracy of 97.00%, a processing time of 5 minutes and 20 seconds, and an F1 score of 0.4888. The study used 27,558 enhanced images, divided into training (22,046) and validation (5,512) sets to avoid overfitting and biases, adhering to an 80% training and 20% validation split. The CNN training model and VGG19 transfer learning achieved an astounding average accuracy of 97.76% for both parasitized and uninfected malaria cell images. The study emphasizes the need for collecting high-quality, abundant image data to effectively analyze contamination levels in both parasitized and uninfected individuals. It is advised that sophisticated microscopic technology be used to improve the performance, accuracy, and speed of malaria diagnosis and prediction utilizing the proposed model.