This paper presents the development and performance evaluation of an Enhanced Chicken Swarm Optimization (ECSO) algorithm for hyperparameter tuning in Convolutional Neural Networks (CNNs), tailored specifically for handwritten document recognition. The ECSO algorithm integrates Gaussian and Tent chaotic maps into the standard Chicken Swarm Optimization framework to enhance the exploration of the solution space and overcome premature convergence. The model was applied to classify original, disguised, and forged handwritten samples using a deep learning framework implemented in MATLAB R2020a. Performance metrics such as accuracy, precision, false positive rate, and execution time were used to assess model effectiveness. Experimental results of the ECSO-CNN model demonstrated strong recognition accuracy (92.14%–95.00%), high precision (95.40%– 97.70%), low false positive rates (4%–8%), and fast execution time (22–28 seconds). These outcomes affirm the robustness and efficiency of the ECSO algorithm in optimizing deep learning architectures for forensic purposes. The study highlights the practical utility of chaos- enhanced metaheuristic algorithms in legal, security, and archival document authentication systems, positioning ECSO-CNN as a viable solution for scalable and intelligent forensic handwriting analysis.
Keywords: ECSO, CNN, Chaotic Map, Handwritten Document Recognition, Hyperparameter Optimization, Forensic AI, Swarm Intelligence
