[3I1-GS-13] AI application: Social application (2)

[3I1-GS-13-05] Finding Common Features Among Multiple Groups in Wagyu Data Analysis

〇Nanami Higashiguchi¹, Masatsugu Motohiro¹, Haruka Ikegami², Tamako Matsuhashi², Kazuya Matsumoto², Takuya Yoshihiro¹ (1. Wakayam University, 2. Kindai University)

Keywords:Wagyu, Carcass Characteristics, Feature Selection, Multi-task LASSO, fairness-index

Wagyu is known in the world as a branded beef of Japan. We are exploring how to predict Wagyu beef quality from protein expression profiles of early-stage beef cattle. Since the protein expression data has a large amount of proteins, we must select a part of them that is truly correlated with beef quality. As the sparse linear regression method, LASSO (Least Absolute Shrinkage and Selection Operator) is the best-known. Although LASSO retrieves a small number of features that explains the target traits, it does not aware groups of samples that has different trends. Unfortunately, it is known that Wagyu data has different trends with each branded region because of the difference in raising methods of Wagyu beef. In this study, we propose a method to select features that commonly effects on beef quality among multiple regions, rather than features specific to each region.