Optimization of Convolutional Neural Network Hyperparameters Using Sinusoidal Chaotic Transit Search for Lung Cancer Identification

Early and accurate lung cancer detection remains a critical clinical priority due to the high mortality associated with late diagnosis. This study presents an optimization framework for convolutional neural network hyperparameters using a sinusoidal chaotic transit search algorithm for lung cancer identification. The model integrates chaotic mapping into the transit search optimization process to enhance global exploration and improve convergence. The system was implemented in MATLAB R2023a and evaluated using publicly available lung cancer datasets from Kaggle and Mendeley. The preprocessing pipeline included Contrast Limited Adaptive Histogram Equalization (CLAHE), grayscale conversion, morphological thinning, and Fuzzy C-means clustering for region-of-interest segmentation. Performance was evaluated using accuracy, sensitivity, specificity, false positive rate, and identification time. Results from the study show that the optimized model outperformed the conventional transit search-based CNN and baseline CNN models with a threshold of 0.75, SCTS-CNN, TS-CNN and Baseline CNN achieved accuracy of 96%, 94% and 89% respectively, specificity of 97%, 95%and 90% respectively and sensitivity of 96%, 93% and 88% respectively for Malignant classification, alongside the lowest FPR of 3%, 5% and 10% respectively and the fastest computation time of 88seconds, 95seonds and 82seconds.. The findings indicate that chaotic enhancement improves feature selection stability and classification accuracy. The study supports the use of metaheuristic-driven deep learning optimization in medical image analysis.