Game-Based Diagnostics for Autism Spectrum Disorder: Integrating Machine Learning with Behavioral and Cognitive Metrics

This study introduces a novel approach to simplifying the diagnosis process of autism by integrating game-based assessments and machine learning technology. We designed an interactive test suite in an easy-to-use, virtual environment for people with autism spectrum disorder (ASD). The system quantifies multiple behavioural metrics: reaction time, task-switching capabilities, and sensory response patterns, while integrating standardized clinical diagnostic data. Our methodology involved 6 male participants aged 15-17: 3 properly diagnosed with autism by official medical professionals (classified as ASD level 1 under testing standards) and 3 neurotypical participants exhibiting normal behavior.Data from gameplay sessions were systematically logged, analyzed, and converted into quantifiable features associated with ASD markers. We evaluated 25 different machine learning classifiers, with AdaBoostClassifier achieving perfect accuracy (100%), while BaggingClassifier and DecisionTreeClassifier provided efficient alternatives at 80% accuracy. Results confirmed established literature on executive function deficits and atypical sensory processing in ASD, though our findings on reaction time contradicted previous research by showing faster responses in ASD participants. This game-based automated diagnostic system demonstrates potential to augment conventional clinical evaluations by offering a low-cost, accessible, and engaging avenue for early ASD detection. While designed as a complementary tool rather than a replacement for formal clinical assessment, our approach addresses significant barriers in current diagnostic practices including resource intensity, time constraints, and specialist dependency. Future improvements will focus on increasing sample diversity, expanding test modules, and refining the user interface to enhance accessibility and reliability across broader populations.