A new Autism Spectrum Disorder Discovery (ASDD) strategy using data mining techniques based on blood tests. (March 2023)

• Record Type: Journal Article
• Authors: Saleh, Ahmed I.; Rabie, Asmaa H.
• Is Part Of: Biomedical signal processing and control. Volume 81(2023)
• Publication Date: 2023-03
• Subjects: Autism spectrum disorder -- Data mining -- Ensemble classification -- Feature selection -- Outlier rejection; Signal processing -- Periodicals; Biomedical engineering -- Periodicals; Signal Processing, Computer-Assisted -- Periodicals; Image Processing, Computer-Assisted -- Periodicals; Biomedical Engineering -- Periodicals; 610.28
• DOI: 10.1016/j.bspc.2022.104419
• Languages: English
• ISSNs: 1746-8094

Abstract

Highlights: New ASDD strategy is provided to give fast diagnosis for autism patients. ASDD composes of two main layers; Pre-processing Layer and Autism Discovery Layer. In PL, feature selection and outlier rejection are used to filter the used data. Hybrid Rejection Technique as rejection method is main contribution of this paper. Experimental results proven that ASDD strategy outperforms the other strategies. Abstract: Autism Spectrum Disorder (ASD) represents a heterogeneous developmental disability characterized by impairments in social communication and behavioral challenges. The problem of ASD begins with childhood and continues into adolescence as well as adulthood. The number of children suffering from ASD is constantly increasing, so a modern method must be found to make a quick and early diagnosis or discovery of ASD patients with high accuracy. Early discovery of ASD patients aims to quickly take the necessary measures with patients and take care of them until their speedy recovery. During this paper, a new Autism Spectrum Disorder Discovery (ASDD) strategy is provided to give a fast and delicate diagnosis for autism patients. The ASDD composes of two main layers, which are; (i) Pre-processing Layer (PL) and (ii) Autism Discovery Layer (ADL). In the PL, two approaches called feature selection and outlier rejection are used to filter the used data from any non-informative data before starting to train the diagnostic model in the ADL for providing delicate diagnosis. Hybrid Rejection Technique (HRT) as a new outlier rejection method is the main contribution of this paper. HRT includes two main stages called Quick Rejection Stage (QRS) as a fast stage and Precise Rejection Stage (PRS) as a delicate stage. QRS uses a standard deviation to quickly remove outliers and then PRS uses a Hybrid Bio-inspired Optimization Method (HBOM) that combines Binary Genetic Algorithm (BGA) and Binary Gray Wolf Optimizer (BGWO) to delicately eliminate the rest of outliers. According to PL in ASDD strategy, Fisher Score (FS) is applied to select the best subset of features and then HRT is used to reject bad training data. Finally, an Ensemble Technique (ET)that includes three classifiers called Naïve Bayes (NB), K-Nearest Neighbors (KNN), and Deep Learning (DL) is used in ADL to diagnose ASD patients based on the filtered data from PL. The ASDD strategy has been compared to many modern strategies using ASD dataset that includes blood tests from children [1] . Experimental results proven that the ASDD strategy outperforms the other strategies in terms of many metrics called accuracy, error, sensitivity, precision, and execution-time with values 0.92, 0.08, 0.85, 0.80, and 2 s.