Enhanced Detection and Adaptive Learning for Autism

Abstract

Autism Spectrum Disorder (ASD) is a neurodevelopmental condition affecting social interaction, communication and behavior, requiring specialized interventions. The proposed framework enhances autism detection through multi-level assessment. It begins with the Social Communication Questionnaire (SCQ), a 40-item test evaluating social abilities of children. If results indicate a high likelihood of autism, the system advances to image-based detection using transfer learning with MobileNetV2. Transfer learning is a machine learning technique where a model trained on a large dataset is adapted for a new, smaller dataset. A pre-trained model like MobileNetV2 has already learned features from large-scale datasets and is fine-tuned to detect autism-related patterns, improving accuracy with limited data. The dataset comprises 3,000 images of both autistic and non-autistic individuals, enabling robust detection. By leveraging transfer learning, the system simplifies the process, reducing training time while maintaining high accuracy. Following detection, the Childhood Autism Rating Scale (CARS) is used to assess autism severity. CARS is a 15-item questionnaire that evaluates behaviors commonly associated with autism, with each item scored on a scale of 1 to 4. This step classifies and helps to determine the appropriate level of intervention. The final stage integrates an adaptive learning system, offering interactive, visually engaging tools, as autistic children respond better to dynamic content. Here, the videos will be about how children interact with society, their day-to-day life activities, and how they react to different situations. This comprehensive approach ensures early, precise autism detection while enhancing learning experiences through personalized teaching methods.