To search, Click below search items.


All Published Papers Search Service


A Comparative Study on Predicting Autism Spectrum Disorders (ASD) Using Gene Expression and Machine Learning


Hala Alshamlan, Hissah AL-Nojaidi, Maraheb AlSuliman, Reham Alabduljabbar


Vol. 20  No. 11  pp. 66-73


Objective: The aim of this study is to identify gene expressions that could have a high potential to be used in predicting autism spectrum disorder (ASD) by using the filter gene selection method with Bayes Networks classifier compared with the wrapper selection method with the SVM algorithm. Bayes Networks classifier was chosen after testing multiple classifiers on the dataset and it has the highest accuracy. Method: The experimental data used in the analysis comprised an autism microarray dataset from the well-known public repository GEO (NCBI) [1]. We have downloaded it after its normalized as part of previous work [4]. The dataset consists of 146 observations (samples) and 9454 genes (features). We applied the Correlation-based (CFS) Attribute Selection filter after that we tested the different classifiers with 10-fold cross-validation to show the highest accuracy of them to be chosen in the proposed model. Result The best accuracy founded when apply Bayes Networks classifier with 10-fold cross-validation on dataset filtered by the CFS feature selection method, which results in 91.7% accuracy. The proposed model has better accuracy when comparing to the filter gene selection methods proposed in the previous work.


Autism Spectrum Disorder (ASD); Filter gene selection; Bayes Networks; CFS; Machine Learning; Weka.