[4I3-OS-1b] AutoML（自動機械学習）

[4I3-OS-1b-03] Automated Feature Construction with Domain Knowledge Using Language Models

〇Yoichi Hirose¹, Shinichi Shirakawa¹ (1. Yokohama National University)

Keywords:Automated Feature Construction, Language Models

Feature construction is effective in improving machine learning, and its automation methods have been developed. However, they do not consider domain knowledge sufficiently, and their execution time depends on the number of columns in tabular datasets because they create candidate features exhaustively. In this study, we propose a method that predicts features from meta-information in a dataset and aims to exploit domain knowledge using a language model. The experimental results show that the performance of the proposed method is comparable to previous studies and that the execution time does not almost no change with the number of columns in tabular datasets.