A FRILLED-LIZARD ALGORITHM-BASED HYPERPARAMETER OPTIMIZATION FRAMEWORK FOR SUPPORT VECTOR MACHINE IN AUTOMATED POULTRY FECAL IMAGE CLASSIFICATION

Abstract

Coccidiosis and Newcastle Disease(NCD) are among the most disastrous poultry illnesses generally, responsible for enormous profitable losses and pitfalls to food security. Accurate, timely, and computationally effective discovery of these conditions remains a persistent challenge for both marketable and smallholder poultry operations. This research showcases the development of a Frilled Lizard Optimization-based Support Vector Machine (FLO-SVM) for the automated detection and classification of Coccidiosis and Newcastle Disease in poultry using fecal image datasets. The FLO algorithm, a novel bio-inspired metaheuristic, was applied to automatically tune three critical SVM hyperparameters: kernel coefficient (γ), the penalty factor (C), and kernel function type. The utilized dataset sourced from kaggle consist 1,400 poultry images, with Coccidiosis represented by 304 images and Newcastle Disease by 376 images. Pre-processing involved conversion, of grayscale discrepancy improvement via CLAHE, image normalization, and principal component analysis (PCA) for feature extraction. The study employed Model evaluation of 10-fold cross-validation across threshold values of 0.3, 0.4, 0.5, and 0.51. The FLO algorithm converged to an optimal configuration of RBF kernel with C = 5.7242 and γ = 0.7922 after 30 iterations, leading to an achievement of a fitness value of 0.9829. The FLO-SVM resulted to peak accuracy of 97.50% for Newcastle Disease and 97.21% for Coccidiosis, respectively, at threshold 0.51, The outcome outperforms the SVM baseline by over 1.5%. Sensitivity exceeded 94%, and false positive rates were reduced to approximately 1.56–1.64% at threshold 0.51. Execution time was greatly decreased, ranging from 30.63 to 38.26 seconds compared to over 58 seconds for the conventional SVM. The output of the result indicate that FLO-SVM provides an accurate, reliable, and computationally efficient framework for poultry disease detection, with significant practical applicability for on-farm diagnostic systems.