AUTISM SPECTRUM DISORDER DETECTION USING MACHINE LEARNING ON COMBINED SCREENING DATA

Abstract

Autism Spectrum Disorder (ASD) is a heterogeneous neurodevelopmental condition associated with differences in social communication and interaction, together with restricted or repetitive patterns of behavior. Because characteristics may become apparent during early childhood while access to formal assessment can be limited, computational screening approaches have attracted increasing research interest [1] [2] [7] [16]. The present study develops a machine-learning framework for ASD screening using a combined age-group questionnaire dataset. The working dataset contained 1,100 records and 21 original columns, with 707 non-ASD and 393 ASD cases. The target variable was encoded as a binary outcome. Missing numerical values were replaced using the median, and missing categorical values were represented as “Unknown.” To reduce target leakage, the rule-derived result field was removed before model development, and the age_desc field was excluded because it duplicated age-group information. Numerical variables were standardized and categorical variables were transformed using one-hot encoding. Seven supervised classifiers—Logistic Regression, Support Vector Machine (SVM), K-Nearest Neighbors (KNN), Random Forest, Neural Network, Decision Tree, and Naive Bayes—were evaluated using a stratified 80:20 train-test split. Performance was assessed using accuracy, precision, sensitivity, specificity, F1-score, area under the ROC curve (AUC), and Matthews correlation coefficient (MCC). Logistic Regression obtained 100.00% on accuracy, precision, sensitivity, specificity, and F1-score, with AUC = 1.000 and MCC = 1.000 on the held-out test set. The corresponding confusion matrix contained 141 true negatives and 79 true positives, with no false positives or false negatives. SVM was the second-highest performing model, with 98.64% accuracy and AUC = 0.999. The results indicate that the combined questionnaire data contain strong predictive patterns; however, the perfect test-set performance should be treated as an experimental result rather than evidence of clinical readiness. Independent validation, repeated cross-validation, and evaluation across age and demographic subgroups are required before practical deployment.