Int J Performability Eng ›› 2023, Vol. 19 ›› Issue (2): 144-154.doi: 10.23940/ijpe.23.02.p7.144154

Previous Articles    

Hybrid Outlier Detection Strategy and Weighted Decision Matrix Ordinal Classifier for CKD Severity Prediction

P. Antony Seba* and J. V. Bibal Benifa   

  1. Indian Institute of Information Technology Kottayam, Kottayam, 686635, India
  • Contact: * E-mail address: sebaantony.phd201002@iiitkottayam.ac.in

Abstract: This article investigates the chronic kidney disease (CKD) dataset to observe meaningful insights through statistical data analysis. It is aimed to introduce a hybrid outlier detection method and a weighted decision matrix (WDM) ordinal classifier for CKD severity prediction. Attention is focused to discover the insights and to draw conclusions for feature extraction, data pre-processing, and feature selection by considering the domain knowledge, data distribution, and relationship among the variables. A hybrid approach is proposed with skewness of each variable. Interquartile range and standard deviation are introduced to handle the outliers, which are detected using univariate analysis. The various ways the values are missing in the training dataset are considered for imputation. Statistical and supervised learning approaches are utilized for selection of optimal features. The proposed outlier detection method identifies 1% of the data instances, which are extreme far points. The proposed WDM ordinal classifier model for predicting the severity of CKD is robust to feature selection, which yields an accuracy of 94.61% using the optimal features given by RFE.

Key words: feature extraction, outliers, imputation, feature selection, WDMl