Presentation | 2022-06-25 Time Series Analysis of Shapley Values in Machine-Learning Regression Kotaro Kuno, Yukari Shirota, |
---|---|
PDF Download Page | PDF download Page Link |
Abstract(in Japanese) | (See Japanese page) |
Abstract(in English) | In regression analysis of machine learning, Lundberg's SHAP and its libraries are widely used and have contributed greatly to the semantic interpretation of regression analysis across various application fields. In this paper, we introduce an approach to apply SHAP to time series analysis. The advantage of SHAP is that it can evaluate the contribution of each explanatory variable to the target variable value, using the characteristic function of each data. If the target value are time series data, even if the same explanatory variable dataset is used, the SHAP values obtained from the regressions become different. By analyzing the time series changes, it is possible to extract the most important explanatory variables at that time. In this paper, we show the evaluation of explanatory variables based on the characteristics of each company by SHAP, using a case study of the stock price recovery rate after the stock price decline by the COVID-19 in global automobile manufacturing industries. |
Keyword(in Japanese) | (See Japanese page) |
Keyword(in English) | Machine Learning / Regression / Shapley values / SHAP / Time series analysis of SHAP distribution |
Paper # | DE2022-1 |
Date of Issue | 2022-06-17 (DE) |
Conference Information | |
Committee | DE |
---|---|
Conference Date | 2022/6/24(2days) |
Place (in Japanese) | (See Japanese page) |
Place (in English) | Musashino University |
Topics (in Japanese) | (See Japanese page) |
Topics (in English) | Social Computing |
Chair | Naofumi Yoshida(Komazawa Univ.) |
Vice Chair | Akiyoshi Matono(AIST) / Yu Suzuki(Gifu Univ.) |
Secretary | Akiyoshi Matono(Kanagawa Inst. of Tech.) / Yu Suzuki(Osaka Univ.) |
Assistant | Ken Honda(Komazawa Univ.) / Hiroki Nomiya(Kyoto Inst. of Tech) |
Paper Information | |
Registration To | Technical Committee on Data Engineering |
---|---|
Language | JPN |
Title (in Japanese) | (See Japanese page) |
Sub Title (in Japanese) | (See Japanese page) |
Title (in English) | Time Series Analysis of Shapley Values in Machine-Learning Regression |
Sub Title (in English) | |
Keyword(1) | Machine Learning |
Keyword(2) | Regression |
Keyword(3) | Shapley values |
Keyword(4) | SHAP |
Keyword(5) | Time series analysis of SHAP distribution |
1st Author's Name | Kotaro Kuno |
1st Author's Affiliation | Gakushuin University(GakushuinUniv) |
2nd Author's Name | Yukari Shirota |
2nd Author's Affiliation | Gakushuin University(GakushuinUniv) |
Date | 2022-06-25 |
Paper # | DE2022-1 |
Volume (vol) | vol.122 |
Number (no) | DE-88 |
Page | pp.pp.1-6(DE), |
#Pages | 6 |
Date of Issue | 2022-06-17 (DE) |