Modeling and Classification Multicollinear Data using Multinomial Ridge Logistic Regression with Wu-Asar Approach to Submit This Sample Abstract
Giatma Dwijuna Ahadi(a*), Ismaini Zain (a), Santi Puteri Rahayu (a)

(a) Department of Statistics, Faculty of Science and Data Analytics, Institut Teknologi Sepuluh Nopember Surabaya, Indonesia
*giatmaahadi.206003[at]mhs.its.ac.id

Abstract

Classification method is an analysis used to classify an object into a certain category based on explanatory variable. Logistics Regression is a classic method that is used to solve classification problems. Multinomial Logistic Regression is a method to find relationships between nominal response variables with one or more predictor variables. Assumptions on logistics are models not contain multicollinearity. Schaefer, Loi & Wolfe introduced the Ridge Logistic Estimator to solve multicollinearity cases in Logistic Regression. Wu & Asar proposed a new method to reduce bias in parameter estimation. The selection of ridge constant values will be resolved using Ridge-Trace, SRW-constants, and Wu-Asar estimator. The performance test of the Wu-Asar ridge value will be applied to the Iris Dataset in R-software. Iris Data consists of three iris species (0:Setosa, 1:Versicolor, 2:Virginica) and four features as predictor. The best ridge-constant value is obtained by examination the smallest standard-error of parameter estimate. The calculation results show that the Wu-Asar approach is the best ridge constant and individual Wald-test shows significant results. Evaluation of classification based on confusion matrix. The performance of the Wu-Asar estimator on Multinomial Ridge Logistic Regression for the Iris dataset shows very good classification results with 98% global accuracy.

Keywords: Classification, Multinomial, Ridge Logistic Regression, Wu-Asar, Confusion Matrix

Topic: MATHEMATICS AND STATISTICS