Classification of Autistic Spectrum Disorder Using Deep Neural Network With Particle Swarm Optimization

1. Introduction:

Autistic Spectrum Disorder (ASD) is not only detrimental to the afflicted individual but also to the patient's family and but also society at large. Several researchers focusing in the area of behavioral science and computational intelligence have recently worked on ASD. They worked intelligent models like ANN and DNN to detect and diagnosis the ASD. One of the psychiatric problems is Autistic Spectrum Disorder (ASD). ASD is a psychological disorder that limits the use of linguistic, communicative, cognitive, capabilities as properly as social competencies and abilities. In recent times, ASD has been studied in the behavioural sciences using intelligent technique based around machine learning to speed up the screening time or to improve sensitivity, accuracy of the diagnosis process (Thabtah, 2017). The ASD is a group of Neuro-developmental disabilities disorders that cannot be cured but may be ameliorated by early detection with the help of Machine Learning (ML). To deal with the ASD problem, this research paper a ML structure used for ASD screening applying a new class of learning called Deep Learning (DL). This research objective to enhance the ASD screening procedure by including a new DL based on Artificial Neural Network (ANN). The ANN based models have showed their advantage in various classification field such as text classification, image classification, and medical diagnosis. Feature selection was a technique used for data pre-processing and it was ideal to implement it earlier than fitting model in machine learning. Selection by mean here was by selecting features and attributes or feature subset in a dataset to be fit into model and tested for performance. In this we used the meta- heuristic search based method called Particle Swarm Optimization (PSO) it is feature Selection technique (FST). The proposed model MLP, DL and PSO-MLP and PSO-DL is helpful for identification and detection of ASD problems and also important features are recognized in this study.

This paper contributes in three ways

• 1.

  It examines the performance of the MLP and DNN classifiers in terms of accuracy, sensitivity, specificity, and F1-score.

• 2.

  The proposed PSO-CBFS has used for feature optimization.

• 3.

  It compares and analyzes the performance of classifiers MLP and DNN classifiers with optimized features by PSO-CBFS.

2. Related Works:

Many researchers have been working in the field of computer-based disease diagnosis systems. The results of the studies show the usefulness of data mining models in ASD-related problems.