An Automated Detection System of Autism Spectrum Disorder Using Optimal Feature Selection-Based Adaptive RNN with Attention Mechanism

The research introduces an automated system for ASD detection through the implementation of an adaptive RNN combined with optimal features selection together with an attention mechanism. The system operates to detect ASD early through a data-driven methodology by analysing behavioural, visual and auditory indicators that mark symptoms of ASD. This solution provides accurate diagnosis and personalized care that doctors can use because it operates across many scales. The system accepts multiple forms of data from facial expressions together with speech patterns and motor behavioural data. A combination of feature selection methods identifies essential data attributes within raw information to decrease complex dimensions that improves system operational speed. The adaptive RNN model extracts temporal relationships through its learning process and the attention mechanism allows the model to select matching features to detect subtle changes in ASD behaviour properly. The framework adapts automatically because it uses available data to refine itself with a purpose of refining detection accuracy. Research outcome shows that adding the attention mechanism enables the model to select important attributes leading to better detection precision. The feature selection system adds to model performance optimization because it enables the model to concentrate on important signals while reducing potential overfitting problems. The system can apply in real time through its diagnostic excellence which minimizes both false positive and negative results. The proposed automated system represents a major step forward for early ASD detection since it delivers clinicians an effective instrument for early intervention. The precise results of this model stem from its adaptive learning system as well as its attention-based focus which leads to better ASD diagnosis and treatment planning outcomes.