Improving Autism Detection Using GridSearchCV for Severity Level Autism in Indonesian Children

This study suggests using facial expression detection and machine learning techniques to evaluate the severity of autism. Facial images of children with autism spectrum disorder are probably included in the data set that was specifically created for autistic children in Indonesia and used in this study. These data contain information on the severity level of autism based on clinical evaluations and observations obtained from professionals and therapists in Indonesia. This data set is essential to ensure that the suggested machine learning techniques can provide models suitable for Indonesia's autistic population, allowing for their application in pertinent clinical settings. In the experiments conducted with hyperparameters, we attained accuracy rates of 90 for KNN, 85 for Random Forest, 81 for Decision Trees, and 80 for Gradient Boosting. Prior to hyperparameter tuning, the accuracy figures stood at 83 for KNN, 73 for Random Forest, 80 for Decision Trees, and 80 for Gradient Boosting. This under scores the significant enhancement in classification accuracy resulting from hyperparameter utilization in this experiment. © 2024 IEEE.