An Explainable AI Approach for Poultry Disease Detection Using Faecal Image Analysis

Abstract

Coccidiosis, Newcastle Disease, and Salmonellosis are among the most damaging diseases affecting poultry farms across sub-Saharan Africa, and smallholder farmers in Katsina State, Nigeria, are particularly vulnerable due to limited access to fast and affordable diagnostic tools. This research presents a novel Explainable Artificial Intelligence (XAI) framework for automated poultry disease detection through faecal image analysis. The proposed pipeline extracts 24 discriminative features from multi-color-space representations (RGB, HSV, LAB) and texture descriptors (LBP, GLCM, wavelet) from faecal images. Four machine learning classifiers Multi-Layer Perceptron (MLP), Random Forest, Gradient Boosting, and Support Vector Machine (SVM) are trained and evaluated on a balanced dataset of 800 samples across four classes: Healthy, Coccidiosis, Newcastle Disease, and Salmonella, based on the PCR-verified dataset by Machuve et al. (2022). The SVM classifier achieved the best performance with a test accuracy of 96.88% and a 5-fold cross-validation accuracy of 98.00%. SHAP (Shapley Additive Explanations) analysis reveals that blue-channel mean, GLCM energy, and LAB A-channel mean are the most discriminative features for disease classification. To address the explainability gap identified in the literature (Abdusalaam et al., 2025; Salih et al., 2025), this study integrates global SHAP explanations with instance-level LIME interpretations, providing veterinary practitioners and farmers with transparent, feature-level reasoning for every prediction. All results, figures, and analyses in this paper are entirely original generated from models trained specifically for this study.