A Hybrid Approach to Support the Detection of Autism Spectrum Disorder(ASD) through Machine Learning and Deep Learning Techniques

Autism spectrum disorder is a significant contributor to developmental disabilities worldwide, and its estimated prevalence in India is more than 2 million people. Research has emphasized that early detection and treatment can help to improve a person with autism's level of language, communication, and overall well-being. The main objective is to automate the diagnosis process, compare and identify the best classifiers, and find out the most significant traits for supporting detection. We attempted to investigate the use of Logistic Regression, Decision Tree, Naive Bayes, Random Forest, Support Vector Machine, Linear Discriminant Analysis, K-Nearest Neighbors, XGBoost Classifier, Multi-Layer Perceptron for predicting if the person is susceptible to ASD. For the analysis, ASD data for toddlers, children, adolescents, and adults were employed, and the best classifier techniques were determined using measures like recall, precision, F-measures, etc. Feature engineering was performed to identify the most impactful and effective features of the datasets. We identified that LR and MLP models provide the best results in adult and toddler datasets. In case of adolescent data, LR classifier showed the highest performance whereas LR and XGBoost algorithms worked well for child dataset. The results obtained justify that ML and DL techniques showed promise in developing an improved system for supporting the detection of Autism.