Comparative Evaluation of Hybrid KNN with Lasso Regression Vs Standalone KNN for Autism Spectrum Disorder Prediction

Autism Spectrum Disorder (ASD) is a multifaceted neurodevelopmental condition that necessitates precise and early identification for timely intervention. Computational learning methods have been extensively investigated for ASD detection, with algorithms such as K-Nearest Neighbors (KNN) being commonly employed due to their straightforward nature and reliability. However, KNN alone may not always provide the best predictive accuracy due to its sensitivity to noise and lack of feature selection capability. This study evaluates a hybrid model combining KNN with Lasso Regression (KNN - Lasso) to enhance predictive performance. Lasso Regression is used for feature selection and regularization, helping to improve model interpretability and reduce overfitting, while KNN classifies individuals based on the refined feature set. We evaluate the effectiveness of the hybrid model in comparison to the standalone KNN based on accuracy, precision, and recall, utilizing ASD screening datasets. Experimental results demonstrate that the hybrid model outperforms KNN by improving classification accuracy and robustness. This study underscores the promise of integrated machine learning approaches for ASD detection, providing a more dependable diagnostic solution for both clinical practice and research applications.