Autism Spectrum Disorder (ASD) is a difficult neurodevelopmental condition categorized with a variety of signs and severity. It encompasses a range of behavioural and social abnormalities, leading to difficulties in societal skills, repetitive behaviors, speech, and nonverbal communication. Although at hand is no definitive cure for ASD, primary finding allows for precautionary measures. Early and precise finding is essential for timely intervention. This review article explores the usage of Machine learning (ML) and Deep learning (DL) methods to develop predictive models for the diagnosis. Traditional diagnostic methods, relying on behavioural observation and interviews, often lack accuracy. Moreover, neuroimaging shows minimal differences between ASD subjects and healthy controls, complicating its use for diagnosis. Consequently, machine learning-based approaches for diagnosing autism are gaining popularity, utilizing features extracted from functional or structural MRI images. Considering the importance of age in structural image studies for diagnosis, we can use the dataset, covering a wide age range. The proposed methods will contribute to building an early diagnosis model for ASD. The aim of the article is to systematically analyses and evaluates recent advancements in using ML and DL techniques for different biomarkers, such as behavior data, genetic, neuroimaging, for the classification and early assessment of ASD. This review explores expansions in machine learning applications in medical imaging, aiming on deep learning techniques, challenges, and future visions.