An adaptive kernel-based weighted extreme learning machine approach for effective detection of Parkinson's disease. (September 2017)

• Record Type: Journal Article
• Authors: Wang, Yang; Wang, An-Na; Ai, Qing; Sun, Hai-Jing
• Is Part Of: Biomedical signal processing and control. Volume 38(2017)
• Page Start: 400
• Page End: 410
• Publication Date: 2017-09
• Subjects: Parkinson's disease -- Imbalanced data -- Extreme learning machine -- Artificial bee colony -- Feature selection; 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.2017.06.015
• Languages: English
• ISSNs: 1746-8094

Abstract

Highlights: An adaptive kernel-based weighted extreme learning machine approach is proposed for Parkinson's disease (PD) diagnosis. Weighted strategy and non-linear mapping of kernel function are used for handling imbalanced data and improving extent of linear separation. Both binary version and continuous version of an adaptive ABC algorithm are used for performing feature selection and parameters optimization. The effectiveness of the proposed method has been evaluated on PD data set in accordance with specificity, sensitivity, ACC, G-mean and F-measure. We have achieved better performance than existing methods in the literature. Abstract: Imbalanced data appear in many real-world applications, from biomedical application to network intrusion or fraud detection, etc. Existing methods for Parkinson's disease (PD) diagnosis are usually more concerned with overall accuracy (ACC), but ignore the classification performance of the minority class. To alleviate the bias against performance caused by imbalanced data, in this paper, an effective method named AABC-KWELM has been proposed for PD detection. First, based on a fast classifier extreme learning machine (ELM), weighted strategy is used for dealing with imbalanced data and non-linear mapping of kernel function is used for improving the extent of linear separation. Furthermore, both binary version and continuous version of an adaptive artificial bee colony (AABC) algorithm are used for performing feature selection and parameters optimization, respectively. Finally, PD data set is used for evaluating rigorously the effectiveness of the proposed method in accordance with specificity, sensitivity, ACC, G-mean and F-measure. Experimental results demonstrate that the proposed AABC-KWELM remarkably outperforms other approaches in the literature and obtains better classification performance via 5-fold cross-validation (CV), with specificity of 100%, sensitivity of 98.62%, ACC of 98.97%, G-mean of 99.30%, and F-measure of 99.30%.