Predicting Autism Spectrum Disorder Using Machine Learning Models

Autism Spectrum Disorder (ASD) is a neurodevelopmental disease that impairs a person’s communication, conduct, and social relationships. Early identification is critical because it increases access to timely therapies, hence improving overall quality of life. Traditional diagnostic approaches, on the other hand, frequently necessitate comprehensive clinical evaluations, making them costly, time-consuming, and difficult to obtain, particularly in underserved areas.To overcome these issues, this research introduces a machine learning-based method that uses automation and data to streamline ASD identification. To improve data quality, the system processes responses from ASD screening questionnaires using techniques such as missing value handling, categorical data encoding, and numerical input normalization. Since ASD cases are frequently underrepresented in datasets, Random Oversampling is employed to balance the dataset, preventing the model from favoring the majority. Feature engineering techniques, such as grouping people into age groups and generating an overall ASD risk score from questionnaire responses, are used to further improve predictions. The approach utilizes numerous classification models, including Logistic Regression, Support Vector Machine (SVM), and XGBoost, to determine ASD risk levels. To guarantee accuracy and dependability, these models are assessed using important performance metrics such as confusion matrices and ROC-AUC scores.Data visualization tools, including as heatmaps, box plots, and count plots, are utilized to get deeper insights into feature correlations and model performance. This approach aims to improve accessibility to early ASD detection by acting as a scalable and affordable screening tool that can be incorporated into digital health platforms.