Dominance Rough Set Theory and Machine Learning based Counterfactual Analysis for Improving Neurodevelopmental Disorder Diagnosis

Abstract

The adoption of Machine Learning into digital platforms is rapidly accelerating progress in many areas, and making it more possible than ever to create efficient and reliable automated healthcare solutions. In the recent past, several approaches have been introduced that use hybrid machine learning models for efficient and accurate Autism Spectrum Disorder detection. However, challenges remain, including limited generalizability and a lack of standardized diagnostic algorithms. This research addresses these problems and recommends an adaptive solution that enhances disease diagnosis and leads to improved healthcare outcomes through a combination of association rules based on machine learning and dominance rough set theory for classifying acute and life-threatening illnesses. Dominance rough set theory is used to select important features to deal with ordinal and preference order information, so more feasible and natural decision-making and discovery of patterns in the multicriteria decision problems can be reached. The experiment is conducted using six well-known machine learning classifiers, namely Association Rules, Decision Tree, Random Forest, K-Nearest Neighbors, Linear Support Vector Machine, and Naive Bayes, on publicly available datasets. A comparison of the performance of different classifiers is made in terms of accuracy. Based on the experimental performance, the results prove that the Association Rules classifier performed better, achieving an accuracy of 99% on the autism adults’ dataset, 97% on the autism toddlers’ dataset, and 97% on the combined autism dataset. Hence, the overall performance strongly indicates that the proposed framework provides accurate, low-cost, and interpretable disease prediction.