Accurate and timely diagnosis of disorder known as autism spectrum disorder (ASD) is not an easy task due to the complicated neurodevelopmental condition's high clinical presentation variation. In order to improve the diagnostic procedure for ASD in pediatric patients, machine learning (ML) techniques have come to light as potential approaches. The previous surveys about the practice of ML algorithms for diagnosing ASD in children has been thoroughly reviewed and summarized. The supervised and unsupervised learning, feature selection, and ensemble methods used in ASD research are among the many ML techniques that is methodically examined. The necessity of large-scale, diverse datasets, cross-validation methods, and interpretability are emphasized over the advantages, disadvantages, and potential future directions of ML-based ASD diagnostic models. This study attempts to offer insights for researchers, clinicians, and other stakeholders in the field of ASD diagnosis by critically analyzing the current status of ML in ASD Diagnosis.