ASSOCIATION RULES BASED CLASSIFICATION FOR AUTISM SPECTRUM DISORDER DETECTION

Autism Spectrum Disorder (ASD) is a very complicated disorder; a recent study made in the USA showed that it is the second most widespread neurodevelopmental disorder among children. It is a highly undetectable disease since its symptoms are not the same for each child. The traditional detecting methods are not efficient especially in the early years of child development, resulting in late treatment. This paper proposes a machine learning approach that detects ASD through gene signature, at childbirth. The approach processes the data of blood-based gene expression of 21 ASD child and 63 Controls into a feature selection phase using Bagging Tree Algorithm, the delivered features are then weighted through the application of a genetic mathematical formula. A Discretization process is used before the main Association Rules process. The final association rules show the most important relationships between genes for the early prediction of the ASD. The rules showed relationships between 600 genes in which 8 genes are the most affecting ASD with an accuracy of 90%.