A Comparative Analysis of Deep Learning Models for Automated Snakebite Wound Classification

Snakebite envenomation remains a severe global health challenge, responsible for millions of cases and over 100,000 deaths each year, particularly in rural and underserved regions. This study compares multiple convolutional neural network (CNN) architectures, including EfficientNet variants (B0, B3, B4, V2B0, V2L), ResNet50, MobileNetV3-Large, and ConvNeXt-Large, for the classification of snakebite wounds. After minimal preprocessing through normalization, data augmentation, and class weighting due to a limited dataset size, the models are evaluated through metrics such as accuracy, precision, recall, F1-score, Matthews Correlation Coefficient, Cohen’s Kappa, and Average Precision, with bootstrap resampling ensuring statistical rigor. EfficientNetV2B0 achieved the highest accuracy, 91.53%, and perfect recall for venomous bites, minimizing life-threatening false negatives. These findings highlight the potential of AI-based snakebite diagnostics to provide immediate, reliable guidance. The discussion emphasizes the clinical implications of this performance, and this study concludes that integrating the top model into a mobile health application could improve survival rates by enabling effective initial care before patients reach a doctor.