Automated Malaria Detection Using EfficientNetB3 on Microscopic Cell Images
Aditya Kumar, Leema Nelson · 2025
Malaria remains a primary global health concern, especially in tropical regions where it still causes high morbidity and death. Early, accurate identification of malaria is essential for effective therapy and disease prevention. This effort aims to design an EfficientNetB3 deep-learning model to automatically detect malaria-infected cells from microscopic images. Using Kaggle's “Malaria Cell Images Dataset,” which has 27,558 images split equally between Parasitised and Uninfected classes, this balanced set is methodically separated into training (80%), validation (10%), and testing (10%) subsets to maximize model development. Comprising 22,506 images, the training data gives the model various instances; the validation data fine-tunes the model to avoid overfitting and assure generalisability. The model's performance is methodically evaluated using the testing data with 2,756 images. With an Fl-score of 96%, recall of 96%, and precision of 97%, the EfficientNetB3 model could correctly distinguish parasitized from uninfected cells, obtaining a high test accuracy of 96.42%. These results show the model's potential as a reliable tool for automated malaria detection, especially in settings with limited resources where early and precise diagnosis is crucial for effective malaria prevention and treatment. This study highlights the significant role of deep learning in advancing global health diagnostics.