The diagnosis of autism spectrum disorder (ASD) has grown more complicated over time with the increased focus and need for health services directed toward it due to its major public health issue. In this study, we established an Outlier-Driven Prediction Framework which ASD detection through hyperparameter optimization combined with outlier-driven machine learning predictive models. We analyzed the performance of various classifiers: Decision Trees, Random Forests (Entropy and Gini), Gaussian Naive Bayes, and Extra Trees, with emphasis on their applicability to high-dimensional, severely imbalanced datasets, which is typical in clinical practice for ASD diagnosis. Outlier robust feature scaling and efficient feature scaling were performed using the local outlier factor (LOF) method. The dataset was retrieved from the UCI ML Repository, and the subjects were adult patients diagnosed with ASD. Using the proposed ensemble classifier method, an accuracy of 98.97% was achieved with the Extra Trees classifier. This framework validates the approach of using ensemble models combined with outlier techniques in ASD diagnostic methods, presenting a novel approach to the integration of machine learning in healthcare.