Deep Learning for Automated Malaria Detection: A EfficientNetB3-Based Analysis of Microscopic Cell Images

This study uses deep learning methods, notably EfficientNet B3, for malaria diagnosis by classifying parasitized and uninfected cell images. This study will design a robust, completely automated system that could support malaria screening to reduce human reviewers and amplify the diagnostic standard. Preprocessing and normalizing the dataset from Kaggle for the model resulted in 27,558 images. Stratified splitting has been done to obtain an equal distribution while dividing the dataset for training, validation, and testing of the model. Feature extraction and classification have been done using transfer learning with an EfficientNetB3 model and additional custom dense layers. The model performed with an accuracy of 95%, while there was consistency in the precision, recall, and f1 scores in both categories. That proves overfitting and generalization strengths. Model evaluation through performance indicators such as the confusion matrix and classification report shows the uncertainty in deploying such a model for real-world use. Recent research has also underlined the recent rapid advance in deep learning for malaria detection and its possible application in low-resource environments. Additional work in data augmentation, in combination with ensemble learning, may further enhance the robustness of the model, increasing its contribution to malaria eradication at a global scale.