Scholarlink

<p>The ability to accurately predict solar radiation is a fundamental requirement for the planning and designing of efficient smart solar energy. Though machine learning can be an effective tool, the black-box principle has made decision-making difficult. In an attempt to address this challenge, we compare the five explainability AI algorithms: SHAP, LIME, ELI5, DALEX, and SHAPASH, as they can be used to effectively predict solar radiation. The algorithms are compared with an XGBoost regression algorithm. The tool participants are compared and evaluated, focusing on how effectively the algorithms can be used to make global and local interpretations, and their usability. Additionally, the study evaluates the algorithms through an ablation test, where the machine learning model is run without some parameters to determine the impact on the overall interpretability. Ultimately, the accuracy of the algorithms is also compared. In all the five algorithms, the temperature is the most dominant factor, followed by other parameters such as humidity and wind. Ultimately, the algorithms are compared, and SHAP and SHAPASH are considered to be the most effective, whereas the simpler interpretations are offered by the ELI5 tool.</p>

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

An empirical analysis of explainable artificial intelligence tool for solar radiation prediction

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年5月25日

查看全文
  • Chowdhury Farjana Tur Santona,
  • Nusrat Sharmin,
  • Muhammad Samee Sevas,
  • Rubyeat Islam,
  • Saidur Rahaman Sagor

摘要

The ability to accurately predict solar radiation is a fundamental requirement for the planning and designing of efficient smart solar energy. Though machine learning can be an effective tool, the black-box principle has made decision-making difficult. In an attempt to address this challenge, we compare the five explainability AI algorithms: SHAP, LIME, ELI5, DALEX, and SHAPASH, as they can be used to effectively predict solar radiation. The algorithms are compared with an XGBoost regression algorithm. The tool participants are compared and evaluated, focusing on how effectively the algorithms can be used to make global and local interpretations, and their usability. Additionally, the study evaluates the algorithms through an ablation test, where the machine learning model is run without some parameters to determine the impact on the overall interpretability. Ultimately, the accuracy of the algorithms is also compared. In all the five algorithms, the temperature is the most dominant factor, followed by other parameters such as humidity and wind. Ultimately, the algorithms are compared, and SHAP and SHAPASH are considered to be the most effective, whereas the simpler interpretations are offered by the ELI5 tool.