- 期刊
Autism Prediction and Identification of Important Features Using Machine Learning Techniques
摘要
Autistic Spectrum Disorder (ASD) is a neurological developmental disorder that causes difficulties in social interaction and communication in an individual. Furthermore, ASD causes patients to exhibit repetitive, constrained patterns of behavior, interests, or activities. Therefore, early prediction of ASD is highly useful for both health care providers and other stakeholders and can help to improve patients' circumstances overall. However, prediction of ASD disease is a critical challenge in clinical data analysis. Traditional methods of detecting autism, such as screening methods employed in well-equipped hospitals, are time consuming and costly. Furthermore, due to costs and screening schedules, early prediction of the disease is not feasible. In this study, we applied multiple machine learning algorithms to predict ASD early and efficiently in three real-world ASD datasets, which include patients of various ages: child, adolescent, and adult. To analyze the effectiveness of the selected algorithms, a comprehensive assessment was performed in terms of accuracy, precision, recall, and F1 score, where the tree-based classifiers (Random Forest and Decision Tree) outperformed others by providing 100 percent accuracy in ASD classification. Moreover, we applied advanced feature important analysis method (SHAP) to investigate the important features in ASD prediction for all three datasets. The findings revealed that the total screening score is the most significant factor in the ASD analysis, regardless of age groups.