A generalized goal programming model for parsimonious robust clusterwise linear regression. Issue 1 (2nd January 2019)

• Record Type: Journal Article
• Authors: Ismail, Eman; Rashwan, Mahmoud; Makary, Nadia
• Is Part Of: Journal of statistics & management systems. Volume 22:Issue 1(2019)
• Page Start: 51
• Page End: 71
• Publication Date: 2019-01-02
• Subjects: (2010) 90C29 -- 62J05 -- 65C05; Goal programming -- Variable selection -- Outlier detection -- Clusterwise regression -- Simulation -- Data analysis; Statistics -- Periodicals; Mathematical models -- Periodicals; Mathematical models; Statistics; Periodicals; 519.5
• DOI: 10.1080/09720510.2018.1522801
• Languages: English
• ISSNs: 0972-0510

Abstract

Abstract: In the last decades, clusterwise regression has received considerable attention. Existence of outliers in the dataset and/or usage of unimportant explanatory variables in fitting the regression lines can lead to inaccurate clustering and regression lines. In this paper, we propose a Mixed Integer Programming (MIP) model to obtain a parsimonious robust clusterwise linear regression. The proposed model also can determine the number of homogenous clusters in the dataset and detect the outliers. The proposed model is applied on some datasets and the results are promising. Also, the performance of the model is evaluated using a simulation study.