A Dissertation on Detection of Autism Spectrum Disorder by Machine Learning Algorithm

Abstract

This work proposes an investigative strategy to examining the efficacy of alternative boosting algorithms in terms of improving the accuracy of diagnosing Autism Spectrum Disorder (ASD). When it comes to autism spectrum disorder (ASD), the average number of cases per 10,000 people has increased from 1.9 in 1980 to 14.8 in 2010, according to data from Asia. Early detection and identification are crucial for improved treatment outcomes in ASD. Different boosting machine learning algorithms have been shown to be an effective technique for detecting autism spectrum disorder (ASD) when it is still in its early stages. This research utilized a dataset containing a total of 1100 instances that was amalgamated from three datasets, with 104 instances being teenagers and 704 instances being adults, with the remainder of the instances being kid instances, from the University of California Irvine (UCI) repository. This dataset was used to train and test the model classification classifier. To begin the functioning of the classifiers, several methodologies were used to build distinct data frames and correlation heatmaps, which were then compared. Eight machine learning methods were investigated, and their performance parameters such as the confusion matrix and accuracy were measured and compared to one another. Furthermore, a comprehensive comparative research was carried out by simulating the precision, sensitivity, F1 score, and ROC-AUC of each algorithm and comparing the results.