CMStatistics 2022: Start Registration
View Submission - CMStatistics
B1615
Title: Two-mode cluster elastic net Authors:  Kaito Oi - Doshisha University (Japan) [presenting]
Shintaro Yuki - Doshisha University (Japan)
Kensuke Tanioka - Doshisha University (Japan)
Hiroshi Yadohisa - Doshisha University (Japan)
Abstract: When multiple regression analysis is applied to data with highly correlated groups of explanatory variables, the estimation of regression coefficients becomes unstable. Cluster elastic net (CEN) has been proposed as one of the methods to solve this problem. CEN infers clusters of features from the data based on the correlation among the variables and association with the response. As a result, CEN can predict the target variable with higher accuracy than multiple regression analysis when applied to such data. However, a drawback of CEN is that the prediction accuracy of the objective variable is reduced for data in which there are several unknown latent homogeneous groups. We propose a two-mode CEN to estimate clusters of individuals and partial regression coefficients for each cluster of individuals that improves the prediction accuracy of the target variable. This method improves the prediction accuracy of the target variable over CEN for data with several unknown latent homogeneous groups. We illustrate the performance of our approach through a simulation study and applying genetic data.