Autism spectrum disorder (ASD) is a neuro-developmental disorder that affects 1% of children and has a lifetime effect on communication and interaction. Early prediction can address this problem by decreasing the severity. This paper presents a deep learning-based transfer learning applied to resting state fMRI images for predicting the autism disorder features. We worked with CNN and different transfer learning models such as Inception-V3, Resnet, Densenet, VGG16, and Mobilenet. We performed extensive experiments and provided a comparative study for different transfer learning models to predict the classification of ASD. Results demonstrated that VGG16 achieves high classification accuracy of 95.8% and outperforms the rest of the transfer learning models proposed in this paper and has an average improvement of 4.96% in terms of accuracy.