Ivanne Dave L. Bayer, Leigjhan Clive B. Lim
The Philippines is home to 147 snake species, 33 of which are dangerously venomous, yet effective identification systems remain lacking. This study proposes an ensemble transfer learning approach for binary classification of Philippine snakes as venomous or non-venomous, aiming to support medical treatment, conservation, and public safety. Snake images were collected from Roboflow's Calabarzon dataset for training and validation, with additional testing data sourced from a Philippine snake identification group on Facebook. Three pretrained convolutional neural networks - ResNet50, VGG16, and DenseNet201 - were fine-tuned as base learners, with outputs combined using XGBoost as a meta-classifier. Preprocessing steps included salient object detection, data augmentation, and normalization to improve model generalization. Experimental results show that while ResNet50, DenseNet201, and VGG16 achieved test accuracies of 83.11%, 80.11%, and 82.22% respectively, the ensemble model outperformed them with 85.11% test accuracy, 86.92% precision, 83.94% recall, and an F1-score of 85.40%. The ensemble reduced both false negatives and false positives, offering a more balanced and reliable classifier compared to individual models. These findings highlight the potential of ensemble deep learning methods for real-world snake identification in the Philippines, with applications in snakebite response, biodiversity research, and public awareness. © 2025 IEEE.
University of San Carlos, Department of Computer, Information Sciences and Mathematics, Cebu City, Philippines
Research at a Glance
Register to unlock
Topics & SDG Alignment
Register to unlock
Collaboration
Register to unlock
Author Profile (Selected)
Register to unlock
References Overview
Register to unlock
Journal & Source
Register to unlock
Metadata & Integrity
Register to unlock