Autism spectrum disorder (ASD) is a neurodevelopmental disorder characterized by difficulties in speech, communication, social interaction, and repetitive behaviors. These challenges hinder language and cognitive development. ASD manifests itself in a wide range of symptoms of varying severity, making it difficult to pinpoint a single cause. Despite numerous efforts, the underlying cause of autism remains unknown. It has been hypothesized that ASD is due to impairment in prediction. The brain constantly processes information and makes predictions about the natural world, which is essential for functioning and habituation. When prediction mechanisms are impaired, the world may feel overwhelming, unpredictable and magical. This has led to the hypothesis about prediction impairment. However, the ability to predict in individuals with ASD has not been extensively studied, particularly across different age groups. In this paper, we developed a machine learning based video game to evaluate the prediction abilities of children with ASD and neurotypical (NT) participants, across two different age groups. We selected subjects from 4 to 7 years and 8 to 12 years, based on the severity of their condition. The study included 12 ASD and 12 NT children aged 4 to 7 years, and 16 ASD and 16 NT children aged 8 to 12 years, all chosen based on specific inclusion and exclusion criteria. Results showed that prediction abilities in the 4-7 age group were significantly lower in ASD children compared to their NT counterparts. However, in the 8-12 age group, ASD children's prediction abilities improved significantly, almost matching NT children in some cases. The study also compared ASD and NT participants within each age group, across various parameters such as success rate, prediction response rate, and accurate response rate. It is observed that the gap between ASD and NT subjects closed significantly for all three parameters with age progression. Also, preliminary usability studies suggest that this tablet-based system has the potential to enhance task performance making it a valuable complementary tool for therapists.