Scholarlink

Based on the mutual information feature selection algorithm, this paper proposes a new feature selection algorithm, Small Sample Exclusion - Minimum Redundancy Maximum Relevance (SSE-mRMR). A temperature estimation experiment is carried out using a real CPU dataset, and the SSE-mRMR algorithm is compared with several currently popular mutual information-based feature selection algorithms, namely MID, MIQ, AMRMR, and SCD. After experimental analysis, the Root Mean Squared Error (RMSE) of the temperature of each CPU core ranges from 1.6 to 2.0, the Mean Absolute Error (MAE) is between 1.0 and 1.2, and the Mean Absolute Percentage Error (MAPE) hovers around 3%. These metrics underscore the algorithm’s efficacy in selecting highly relevant features for regression tasks, with potential applications in robotic systems reliant on thermal management or sensor data optimization.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

SSE-mRMR: A New Feature Selection Method Based on Mutual Information

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Hao Wu,
  • Hai Xiong,
  • Long Cheng

摘要

Based on the mutual information feature selection algorithm, this paper proposes a new feature selection algorithm, Small Sample Exclusion - Minimum Redundancy Maximum Relevance (SSE-mRMR). A temperature estimation experiment is carried out using a real CPU dataset, and the SSE-mRMR algorithm is compared with several currently popular mutual information-based feature selection algorithms, namely MID, MIQ, AMRMR, and SCD. After experimental analysis, the Root Mean Squared Error (RMSE) of the temperature of each CPU core ranges from 1.6 to 2.0, the Mean Absolute Error (MAE) is between 1.0 and 1.2, and the Mean Absolute Percentage Error (MAPE) hovers around 3%. These metrics underscore the algorithm’s efficacy in selecting highly relevant features for regression tasks, with potential applications in robotic systems reliant on thermal management or sensor data optimization.