The Effect of Regularization on Deep Learning Methods For Detection of Malaria Infection

Malaria is an infectious disease caused by peripheral blood parasites of the genus Plasmodium. The estimated global malaria cases reached 229 million cases in 2019, of which 250,644 cases occurred in Indonesia. The large number of cases makes the early and accurate diagnosis of malaria very important because it can reduce the severity and prevent death. The most widely used method of diagnosis by far is examining a thin blood smear under a microscope and looking for infected cells. This research examines the effect of regularization applied to several Convolutional Neural Network (CNN) architectures to obtain the best accuracy of malaria parasite detection on thin blood smear images. The regularization techniques used are dropout layer, L2 regularization, and data augmentation. The results show that the use of BaselineNet without regularization achieved 94.92% accuracy. The use of regularization on ResNet-50, MicroVGGNet, BaselineNet-1 obtained 97.12%, 95.64% and 96,28% accuracy respectively.