Efficiency and Comparative Performance of LBP-Based Random Forest and SVM for Toraja Buffalo Classification

Abstrak

Toraja buffalo holds significant cultural and economic value, yet automated classification remains challenging due to subtle visual differences between types. This study evaluates the efficiency of Local Binary Pattern (LBP) combined with Random Forest (RF) and Support Vector Machine (SVM) for classifying six Toraja buffalo types: Balian, Lotong Boko, Pudu, Saleko, Todi, and Ulu. Unlike complex deep learning approaches, this research focuses on a computationally efficient framework by extracting texture features from multiple body parts: head, eyes, horns, body, and tail to capture distinctive patterns. The methodology involves multi-part feature fusion and a comparative analysis of ensemble versus kernel-based learners. Experimental results demonstrate that the LBP-Random Forest model significantly outperforms SVM, achieving a superior accuracy of 92.08% compared to 64.17%. The findings highlight that the proposed LBP-RF integration provides a robust and resource-efficient alternative for livestock image classification, balancing high diagnostic accuracy with lower computational requirements.