Implementation of EfficientNet-B0 Architecture in Malaria Detection System Based on Patient Red Blood Cell (RBC) Images

Malaria is an infectious disease with the most significant number of sufferers currently. The Plasmodium parasite and diagnosis cause this disease involves observing the patient’s red blood cells (RBC) by medical personnel. However, with technological advances, RBC observation can become easier with the help of artificial intelligence algorithms. Therefore, researchers attempted to develop a malaria detection system based on RBC images using the CNN EfficientNet-B0 method in this study. The RBC dataset was obtained from the publicly accessible National Institutes of Health (NIH) repository. Data pre-processing starts with augmentation, resizing, rotation, and horizontal flip in the initial stage. Then, class weighting is carried out, and the dataset is divided into training, validation, and testing data. In the model training process, $\mathbf{1 0}$-fold cross-validation was used with 15 epochs. The testing results showed that the EfficientNet-B0 model had accuracy, precision, specificity, sensitivity, and F-1 scores, reaching 97.37%, 98.17%, 98.17%, 96.58%, and 97.37%. In addition, at the 12 th epoch, the EfficientNet-B0 model achieved optimal accuracy in the training and validation process. These results are higher and more efficient than other CNN models used in this research, such as VGG19, MobileNet, Inception, and Xception, whose accuracies reached 96.15%, 95.74%, 95.92%, and 95.83%, respectively.