Autism Disease Prediction Using Deep Learning & Transfer Learning

The neurological disease known as autism spectrum disorder (ASD) is characterized by impaired social interaction, communication issues, and constrained and repetitive behavior patterns. For the benefit of early interventions and support for afflicted persons, timely and accurate ASD prognosis is essential. Deep learning methods have become effective tools for predictive modeling across a range of industries, including healthcare. This study examines the use of deep learning and transfer learning to forecast ASD using a large dataset of clinical and behavioral variables. In this study, the effectiveness of three well-known deep learning architectures VGG16, DenseNet121, and MobileNetv2 in predicting ASDs is compared. A sizable dataset with a variety of ASD-related variables, such as demographic data, medical histories, and behavioral assessments, is used to train the models. To take use of pre-learned weights from models trained on extensive generic image recognition tasks, transfer learning is used. With accuracy rates of 97% apiece, the experimental results show remarkable prediction performance for VGG16 and DenseNet121. These models have significant generalization abilities that make it possible to make reliable predictions for identifying those who are at risk for ASD. In contrast to the other architectures, MobileNetv2 only obtains an accuracy of 73%. The results show that deeper architectures like VGG16 and DenseNet121 capture the rich patterns and fine details of the input data, resulting in more precise predictions. Additionally, thorough investigations are carried out to look into the models' learned representations and pinpoint the primary features that influence ASD prediction. These revelations aid in a better comprehension of the underlying causes and potential biomarkers of ASD. The information gleaned from these studies can direct ongoing research projects and support the creation of individualized interventions and therapies. Overall, the study emphasizes how well deep learning and transfer learning methodologies work to forecast ASD. The VGG16 and DenseNet121 high accuracy results show their potential as useful tools for early identification and intervention. This study adds to the body of information on ASD prediction and emphasizes the significance of utilizing cutting-edge machine-learning algorithms to improve clinical judgment and patient outcomes.