Nowadays, Autism Spectrum Disorder (ASD) is one of the primary psychiatric disorders illness that rapidly increases. One of the main problems of medical diagnosis data and classification is the variance in symptoms between patients. Thus, finding the discriminative symptoms that distinguish the illness accurately is an important issue. This paper will explore various feature selection methods on four ASD datasets for extracting significant features for improving the ASD classification system. Datasets were created in 2017 and 2018 for child and adult gathered online. Several feature engineering techniques are applied to rank significant features. The correlation matrix method showed the association between features that enable us to select the highest significant features. Then each dataset split into 70% for training and 30% for test. Several machine learning classifiers are applied. After testing, the selected features achieve 100% accuracy, specificity, sensitivity, AUC, and f1 score with adaboost, linear discriminant analysis and logistic regression classifier on different size of data. I choose the adaboost model because it does the same performance with less time and less computational power in both dataset 2017 and 2018 for child and adult. Results were validated using cross-validation with 10 k-fold. The code applied in that paper in https://github.com/BasmaRG/ASD/ .