A Deep Learning-Based Neural Network Model for Autism Spectrum Disorder Prediction

The autism spectrum disorder (ASD) is a neuro-disorder that tremendously impacts people’s lives and today ASD is gaining its prevalence globally faster than ever. ASD affects the mental, social, and physical state of a person due to its unknown etiology, and medical professionals believe that identifying autistic traits and providing accurate analysis and early ASD detection is a relatively challenging and time-consuming task. However, diagnostic predictions for autism features could be improved using multiple methods owing to the rise and development of artificial intelligence (AI) and machine learning (ML) techniques. Therefore, this research attempts to explore the possibility of using AI deep learning techniques to assist in ASD diagnosis and prediction by proposing an effective prediction models based on deep learning Artificial Neural Networks (ANNs) and Convolutional Neural Network (CNNs). This will aid medical, healthcare professionals and recent research studies in diagnoses of autism using a publicly available dataset. We employed extensive exploratory data analysis and oversampling techniques on the dataset to avoid overfitting problems. We enhanced our deep learning neural network models by selecting the best parameter values and using a validation approach. Our proposed ANN model was able to achieve 92.6% in terms of precision and 88.2% in terms of accuracy, while our CNN model was able to achieve 86.4% in terms of precision and 90.6% in terms of accuracy. Our obtained results are compared and benchmarked against the existing state-of-the-art literature that addresses the same problem.