Classification of Starling Images Using a Bayesian Network

Abstract

The classification of starling species is vital for biodiversity conservation, especially as some species are endangered. This research investigates the effectiveness of the Bayesian Network (BayesNet) for classifying starling species and compares its performance with Artificial Neural Networks (ANN) and Naive Bayes models. The dataset comprises 300 images of five starling species—Bali, Rio, Moon, Kebo, and Uret—captured under controlled conditions. Feature extraction focused on color, texture, and shape, while data augmentation through slight image rotations was applied to enhance model generalization. The BayesNet model achieved an accuracy of 96.29% using a 90:10 training-to-testing split, outperforming ANN (90.74%) and Naive Bayes variants. Precision, recall, F1-score, and AUC-ROC values further validated the robustness of the BayesNet model, with precision at 0.90, recall at 0.91, F1-score at 0.92, and AUC-ROC at 0.95. These results demonstrate the superior performance of multi-feature Bayesian Networks in starling classification compared to other machine learning models. The novelty of this study lies in its application of a probabilistic approach using Bayesian Networks, which enhances interpretability and performance, especially in scenarios with limited data. Future work may explore additional feature sets and advanced machine learning models to further improve classification accuracy and robustness.