Deep Learning-Based Image Classification for Malaria Diagnostics: Comparative Analyses of VGG16 and VGG19 AI Models
Itunuoluwa Isewon, Faith Adegoke, Jelili Oyelade, Joshua Olusegun Okeniyi, Christopher Chintua Enweremadu · 2026
This study employs comparative statistical analyses to investigate the performances of two Visual Geometric Groups, the VGG16 and the VGG19 artificial intelligence (machine learning) models, for deep learning-image based classifications of the microscopy datasets of blood film images for Plasmodium falciparum induced malaria infection. The two deep learning models were separately applied to 27,558 datasets of malaria images obtained from infected and uninfected subjects. Data preprocessing, augmentation with a Python-based Machine Learning implementation via Scikit- learn and Tensor Flow libraries were employed for the feature learning and classification of the medical imaging data, from which selected evaluation metrics were used to evaluate the performances of the two AI models. Model performances were analyzed using the Shapiro-Wilk, Student&s;s t , and Cohen&s;s d statistics. Results showed that the VGG16, though precise, misses substantial portions of actual positive cases (low recall), making it less reliable for malaria diagnostic tasks, whereas the VGG19 exhibited clear superiority across all evaluation metrics of accuracy, precision, recall, F1 and the Matthews Correlation Coefficient (MCC). The Shapiro Wilk tests indicates the performances meet normality assumptions. The Student&s;s t - test&s;s p -values of 5.23 × 10 −12 (accuracy) and 8.55 × 10 −12 (MCC score) exhibited Cohen&s;s d effect sizes of –9.5569 (accuracy) and –9.5569 (MCC score), thus confirming that the differences encountered in the performances of the two AI models are not due to random chance but rather indicate significant variations in their capabilities. The study further emphasize needs for additional optimization using hyperparameter tuning and other deep learning models for enhancing the performances and the use of other Convolutional Neural Network (CNN) architecture.