Comparative Analysis of Automated Malaria Cell Classification: EfficientNet-B0 Transfer Learning Versus Traditional Machine Learning
Mainak Saha, Rabin Bhaumik, Tannistha Pal · Auerbach Publications eBooks · 2026
Manual microscopy remains the gold standard for malaria diagnosis but is time-consuming, labor-intensive, and susceptible to human error, particularly in resource-limited environments. This chapter presents a comparative evaluation of automated malaria cell detection using both traditional machine learning (ML) approaches and a modern deep learning method based on transfer learning. Handcrafted feature-based classifiers, including random forest, k-nearest neighbors (KNN), and support vector machines (SVM), were trained using color, texture, and shape descriptors extracted from blood smear images. In parallel, the EfficientNet-B0 convolutional neural network (CNN) was employed using transfer learning, initialized with ImageNet weights and fine-tuned on 27,558 freely accessible, labeled red blood cell image datasets. The EfficientNet-B0 model achieved a peak validation accuracy of 96.5%, outperforming traditional ML classifiers based on F1 score, recall, and precision. Grad-CAM visualization further revealed that the CNN focused on biologically relevant parasite regions. Our findings demonstrate that transfer learning enables high diagnostic accuracy with minimal training time and no need for manual feature engineering. This study highlights the growing potential of deep learning models for use in therapeutic settings for automated malaria diagnosis and supports their adoption in real-world healthcare scenarios, particularly in malaria-endemic and low-resource regions where rapid, accurate screening is critical.