A Methodical Machine Learning Approach for Autism Stage Classification

Abstract

Since autism is thought to present differently in each person, it is commonly viewed as a spectrum condition. Some, but not all, of these features are shared by people with autism, and an individual's level of autism-related symptoms may also vary. While some autistic persons have no spoken language at all, others are able to communicate normally. There is a wide range in how much assistance an individual need, and even the same individual may appear in quite different ways at various times. Recent versions of the leading diagnostic manuals identify ASD as a single diagnosis, despite the fact that autism was formerly separated into subtypes that have since been questioned for their validity. Overall and by subgroups, this study examined whether the rate of new autism diagnoses has been rising over the past two decades. Logistic Regression (LR), Random Forest (RF), Decision Tree (DT), Gradient Boosting (GB), X-Gaussian Boosting (XGB), k-Nearest Neighbors (KNN), Adaboost Classifier (ADB), and Support Vector Machine (SVM) classifier were used to achieve a maximum classification accuracy of 99.68% in this study. The data was cleaned up and prepared for analysis, and then the ML algorithm was used. The results of each algorithm were compared, and an accurate result was found. The F1 score, together with precision, recall, and the AUG score, is used to evaluate performance. Analyses show that Svm classifier and Gradient Boosting are the most effective methods for reaching an overall accuracy of 99.75%. Keyword Logistic Regression Classifier (LRC), Random Forest Classifier (RFC), Decision Tree, Gradient Boosting, Support Vector machines, K-Nearest Neighbors Classifier(KNNC), Adaboost classification, Gaussian training, XGB classifier.