Exploring Sparse Gaussian Processes for Bayesian Optimization in Convolutional Neural Networks for Autism Classification
Suresh Cheekaty, G. Muneeswari
Abstract
Autism Spectrum Disorder (ASD) constitutes a prevalent childhood condition, impacting approximately 0.62% of the global population. Children grappling with ASD encounter challenges in acquiring language proficiency, comprehending spoken communication, and interpreting nonverbal cues, including eye contact, facial expressions, and hand gestures. This discourse delves into the application of eye tracking as a foundational instrument in ASD screening, leveraging the distinctive attributes of gaze behavior. Within the realm of Artificial Intelligence(AI), Deep Learning(DL) manifests substantial promise, exhibiting considerable strides across diverse domains and emerging as an indispensable technological paradigm. This treatise introduces an innovative Convolutional Neural Network (CNN) model predicated on Bayesian optimization for the discernment of eye-tracking images. The model comprises two primary components: the initial segment employs CNN to extract and amalgamate salient features, while the subsequent section incorporates a Bayesian optimizer tasked with fine-tuning the CNN’s hyperparameters contingent on an objective function. In the preliminary analysis, three distinct scenarios devoid of specific elements are juxtaposed with Bayesian optimization. These scenarios undergo evaluation through convergence charts and accuracy metrics, culminating in an impressive 97% accuracy rate. The proposed algorithm evinces superior performance, yielding a noteworthy accuracy level of 97% in the model evaluation. This outcome constitutes a commendable advancement in research, accentuating the efficacy and resilience of the developed approach. The article undertakes a comparative analysis of research methodologies and thematic analysis approaches, furnishing valuable insights for qualitative researchers and offering methodologically robust pathways to address research inquiries. In practical applications, the model attests to heightened reliability and precision, thereby underscored as efficacious in real-world scenarios.