Early detection of autism spectrum disorder using explainable AI and optimized teaching strategies

Autism spectrum disorder (ASD) is defined by the deficits of social relating, language, object use and understanding, intelligence and learning, and verbal and nonverbal communication. Most of the individuals with ASD have genetic conditions; however, early identification and intervention reduce the use of health services and other diagnostic procedures. The varied nature of ASD is widely acknowledged, with each affected individual displaying distinct traits. The variability among autistic children underscores the challenge of identifying effective teaching strategies, as what works for one child may not be suitable for another. In this study, we merge two ASD screening datasets focusing on toddlers. We employ three feature engineering techniques to extract significant features from the dataset to enhance model performance. This study presents an innovative two-phase method where initially, we employ diverse machine learning models, such as a combination of logistic regression and support vector machine classifiers. The focus of the second phase is on identifying tailored educational methods for children with ASD through the assessment of their behavioral, verbal, and physical responses. The main goal of this study is to develop personalized educational strategies for individuals with ASD. This will be achieved by employing machine learning techniques to enhance precision and better meet their unique needs. Experimental results achieve a classification accuracy of 94% in ASD identification using Chi-square extracted features. Concerning the choice of the best teaching approach for ASD children, the proposed approach shows 99.29% accuracy. Performance comparison with existing studies shows the superior performance of the proposed LR-SVM ensemble coupled with Chi-square features. In conclusion, the proposed approach provides a two-phase strategy for identifying ASD children and offering a suitable teaching strategy with respect to the severity of the ASD, thereby potentially contributing to the development of tailored solutions for children with varying needs. •To promote diversity within the ASD data, new dataset is generated by combining two datasets.•An ensemble model of LR and SVM using soft voting.•Bidirectional elimination, principal component analysis (PCA), and Chi-square (CHI2) are investigated.•Evaluation ASD children concerning verbal, behavioral, and physical abilities.