Some Hybrid Approaches to Prediction of a Binary Outcome

• Activity Number: 554
• Type: Contributed
• Date/Time: Wednesday, August 12, 2015 : 10:30 AM to 12:20 PM
• Sponsor: Section on Statistics in Epidemiology
• Author(s): Ho-Lan Peng* and Chih-Hsien Wu and Wenyaw Chan
• Companies: The University of Texas Health Science Center and and The University of Texas School of Public Health
• Keywords: prediction ; binary outcome ; logistic regression ; k-nearest-neighbor

Abstract

Prediction of the distribution for a binary outcome has been a very important topic in several aspects of the science. For example, prediction of a disease occurrence is a common topic in epidemiology. Among the existing methods, logistic regression and k-nearest-neighbor (KNN) algorithm are most commonly used for carrying out this statistical task. Although KNN is a very efficient procedure, the logistic regression approach can take the advantage of the information provided by the covariates to make a more accurate prediction of the outcome distribution. To the best of our knowledge, no research has studied the combination of these two popular procedures. In this research, we propose a method that combines KNN with different types of binary outcome regression models to examine the accuracy of the proposed methods by comparing with other methods available in the literature. We conduct a simulation study to assess the accuracy of the proposed methods and other available approaches in various scenarios of parameter combinations including the size of the neighbor.