Deep learning-based stacking ensemble for malaria parasite classification in blood smear images

Abstract

Malaria remains a significant global health challenge, necessitating accurate and efficient diagnostic tools. Deep learning models have emerged as promising solutions for automated malaria detection using microscopic blood smear images. This study evaluates the performance of various convolutional neural network (CNN) architectures, including VGG16, ResNet50, MobileNetV2, and EfficientNet, in classifying infected and uninfected cells. Individual model performances were assessed based on accuracy, precision, recall, and F1-score, with EfficientNet achieving the highest standalone accuracy of 88.0%. To enhance classification performance, a stacking ensemble approach was implemented, using a logistic regression meta-classifier to integrate outputs from multiple models for improved decision-making. The stacking model outperformed individual networks, achieving an accuracy of 89.4%, with precision, recall, and F1- scores surpassing those of standalone models. Challenges in malaria parasite classification—such as high inter-class similarity, variations in staining quality, and class imbalance were addressed through data augmentation and model tuning. These findings highlight the potential of ensemble learning in medical image analysis, paving the way for more accurate and scalable malaria detection systems.