Autism spectrum disorder (ASD) is a lifelong neuro-developmental disorder that is generally marked by a set of communication and social impairments. The early diagnosis of autism is genuinely beneficial for the welfare of children and parents as well. However, making an accurate diagnosis of autism remains a challenging task, which requires an intensive clinical assessment. The lack of a gold standard test calls for developing assistive instruments to support the process of examination and diagnosis. In this respect, this chapter seeks to provide practical applications of machine learning (ML) for that purpose. The study stemmed from an interdisciplinary collaboration by joint efforts of psychology and artificial intelligence researchers. The chapter is structured into two main parts as follows. Initially, the first part provides a review of the literature that approached the ASD diagnosis using a variety of ML approaches. Subsequently, the chapter presents a set of empirical ML experiments using an eye-tracking dataset. A vision-based approach is adopted based on the visual representation of eye-tracking scanpaths as a form for learning the behavioral patterns of gaze. The ML experiments include the application of supervised and unsupervised learning. It is practically demonstrated how ML could effectively support the ASD diagnosis through providing a data-driven second opinion.