<![CDATA[Malaria remains an infectious disease that requires a fast and accurate diagnostic process, particularly in areas with limited access to skilled microscopists. Microscopic examination as a conventional diagnostic method still has several limitations, including dependence on the expertise of laboratory analysts, relatively long examination time, and the potential for interpretation errors. This study aims to develop a malaria blood cell image classification system using a deep learning-based transfer learning approach with the ResNet50V2 architecture and to compare its performance with MobileNetV2 as the baseline model. The dataset used in this study was the NIH Malaria Cell Images Dataset, consisting of 27,558 red blood cell images categorized into two classes: Parasitized and Uninfected. The research stages included image preprocessing, data augmentation, dataset splitting with a 70:15:15 ratio, model training, performance evaluation, and implementation of the best-performing model into a web-based application. The evaluation was conducted using accuracy, precision, recall, F1-score, confusion matrix, and ROC-AUC metrics. The results showed that ResNet50V2 achieved an accuracy of 93.76%, precision of 93.81%, recall of 93.76%, F1-score of 93.76%, and an AUC of 0.985. Meanwhile, MobileNetV2 obtained an accuracy of 93.06%, precision of 93.22%, recall of 93.05%, F1-score of 93.05%, and an AUC of 0.980. ResNet50V2 also produced a lower number of false negatives than MobileNetV2, making it more suitable for supporting the detection of malaria-infected blood cell images. The best-performing model was then implemented into a web-based application using Flask API. The findings indicate that ResNet50V2 has the potential to be used as a decision support system for early malaria screening based on digital images, although further clinical validation is still required before implementation in real medical environments.]]>
