The prompt identification and diagnosis of autism spectrum disorder (ASD) is essential for timely intervention. In this study, we investigate the usefulness of Bernoulli’s Naive Bayes algorithm for ASD screening based on demographic and medical features. The majority of the existing literature on ASD is addressed by a prediction system based on typical machine learning methods such as the support vector machine and the random forest algorithm. The primary goal of this research is to create a strengthened framework for ASD prediction researchers. Cross-validation using GridSearchCV and hyperparameter tweaking techniques were used to evaluate a variety of machine learning (ML) models, including Bernoulli Naive Bayes, Random Forest, Support Vector Machine, and others. As a result of this study, an ASD prediction model emerged employing Bernoulli’s Naive Bayes technique. We statistically prove why we used Bernoulli’s Naive Bayes algorithm and not any other ML models. The Bernoulli Naive Bayes classifier has the most outstanding final tune cross-validation accuracy of 92.22%, exhibiting resilience against overfitting the imbalanced dataset. In comparison, the K-Nearest Neighbor performed the poorest, with an accuracy of 87.65%. Our findings demonstrate that Bernoulli Naive Bayes’ strong performance implies that machine learning to screening approaches merit more real-world testing and development.