Mutual Information and Conditional Independent Testing for Causal Feature Selection

• Item type: Symposium(1)
• 公開日: 2022-01-28
• タイトル: Mutual Information and Conditional Independent Testing for Causal Feature Selection
• タイトル / 言語: en
• タイトル: Mutual Information and Conditional Independent Testing for Causal Feature Selection
• 言語: eng
• 資源タイプ / 資源タイプ識別子: http://purl.org/coar/resource_type/c_5794
• 資源タイプ: conference paper
• 著者所属: Department of Electrical Engineering, Faculty of Engineering, Prince of Songkla University
• 著者所属: Department of Electrical Engineering, Faculty of Engineering, Prince of Songkla University
• 著者所属(英): en
• 著者所属(英): Department of Electrical Engineering, Faculty of Engineering, Prince of Songkla University
• 著者所属(英): en
• 著者所属(英): Department of Electrical Engineering, Faculty of Engineering, Prince of Songkla University
• 著者名: Rakkrit, Duangsoithong; Yuying, Zhao
• 著者名(英): Rakkrit, Duangsoithong; Yuying, Zhao
• 論文抄録 / 内容記述タイプ: Other
• 論文抄録 / 内容記述: Causal feature selection algorithms can discover causal relationship; however, the redundant features are difficult to defined in the causal graph. To overcome redundancy analysis in this causal feature selection problem, this paper proposes mutual information and conditional independent testing for causal feature selection algorithm (MICI). According to the results, MICI can remove both irrelevant and redundant features while also discover the causality graph. The average accuracy of MICI slightly improves compared to original data and other feature selection methods.
• 論文抄録(英) / 内容記述タイプ: Other
• 論文抄録(英) / 内容記述: Causal feature selection algorithms can discover causal relationship; however, the redundant features are difficult to defined in the causal graph. To overcome redundancy analysis in this causal feature selection problem, this paper proposes mutual information and conditional independent testing for causal feature selection algorithm (MICI). According to the results, MICI can remove both irrelevant and redundant features while also discover the causality graph. The average accuracy of MICI slightly improves compared to original data and other feature selection methods.
• 書誌情報: Proceedings of Asia Pacific Conference on Robot IoT System Development and Platform; 巻 2021, p. 82-83, 発行日 2022-01-28
• 出版者 / 言語: ja
• 出版者: 情報処理学会