A COMPARATIVE STUDY OF DEEP LEARNING AND TRANSFER LEARNING FOR MALARIA PARASITE CLASSIFICATION

ABSTRACT

Malaria parasite poses a significant threat to the African continent. Despite the investments by governments and international donor organisations the Malaria disease still shows no signs of stopping. Although laboratory methods have yielded positive outcomes, there is still a need to explore other possibilities. In this study, we carried out malaria parasite detection by examination of images of blood samples. This study set out to detect Malaria parasite in blood cells using Convolutional Neural Networks, VGG16, and ResNet with the aid of an image dataset containing 27,558 images from the NIH repository. The CNN model employed a 7-fold validation technique, while the ResNet model used augmentation. Our models performance was calculated using classification metrics. Overall, an accuracy of 94% was achieved using CNN, while VGG16, and ResNet achieved 89% and 67% respectively. The study concludes that CNN is an effective model for malaria parasite detection but recommends that more work is needed for the identification of malaria plus count.

Keywords: Malaria, CNN, Transfer learning, Machine Learning, Deep learning.