COMPARATIVE PERFOMANCE ANALYSIS OF SELECTED CHAOTIC MAPS TO ENHANCE CHICKEN SWARM OPTIMIZATION FOR HANDWRITTEN DOCUMENT

Handwritten document identification (HDI) is a fundamental task in forensic, financial, and archival applications, but intra-writer variability and inter-writer similarity make it difficult to select feature subsets that generalise well, and conventional Chicken Swarm Optimisation (CSO) is prone to premature convergence when applied to high-dimensional feature-selection problems. This study integrated chaotic maps into CSO to improve population diversity, delay premature convergence, and improve writer-identification accuracy. The aim is to evaluate three chaos-enhanced CSO (CE-CSO) variants, Logistic+CSO, Sine+CSO and Tent+CSO, for feature selection on nine geometric, textural, and deep-learning features extracted from 500 handwriting samples belonging to three writers in the IAM Handwriting Database. Chaotic sequences generated by the Logistic, Sine, and Tent maps were substituted for the pseudo-random terms used in population initialisation and in the rooster, hen, and chick position-update rules. The three variants were implemented in Python and evaluated using accuracy, precision, recall, and F1-score, with a Support Vector Machine (RBF kernel) as the fitness evaluator, averaged over 30 independent runs per algorithm. Tent+CSO achieved the best results, with an accuracy of 92.15%, precision of 92.03%, recall of 92.15%, and F1-score of 92.09%, followed by Sine+CSO (90.78%, 90.65%, 90.78%, 90.71%) and Logistic+CSO (89.23%, 89.01%, 89.23%, 89.12%). Tent+CSO also converged fastest, reaching its solution in an average of 36.8 iterations, a 13.4% improvement over Logistic+CSO (42.5 iterations), and retained the smallest feature subset (47.8% reduction), though at the longest mean execution time (49.62 s) of the three variants. The study establishes that Tent+CSO offers the best accuracy-convergence trade-off for this task, Sine+CSO offers an intermediate balance, and Logistic+CSO offers a computationally simpler reference implementation.