Scholarlink

<p>Relative Humidity (RH) modeling is essential in many fields like meteorology, agriculture, and public health. Machine Learning (ML) techniques have become valuable tools for RH estimation. However, many of these algorithms are hard to interpret and trust due to their black-box nature. This study investigates the potential of eXplainable Artificial Intelligence (XAI) methods to elucidate RH predictions made by a Gradient Boosting (GB) model for Fez, Morocco. RH was estimated via seven input variables: air temperature (T), solar radiation (H), dew temperature (Td), sunshine duration (n), pressure (P), dew temperature (Td), precipitation (Pr), and wind speed (v). T and Td were identified as the most significant predictors using Permutation Feature Importance (PFI), Shapley Additive Global Importance (SAGE), t-statistic-based surrogate modeling, and SHapley Additive exPlanations (SHAP). SHAP methods revealed also that T and Td are the most interacting predictors. The GB model, incorporating only these two features, outperformed (R<sup>2</sup> = 0.986, RMSE = 2.166, MAE = 1.621, MAPE = 3.441%) the model utilizing all features (model (R<sup>2</sup> = 0.974, RMSE = 2.633, MAE = 2.045, and MAPE = 4.237%). Partial Dependence Plot (PDP), Individual Conditional Expectation (ICE), Accumulated Local Effects (ALE), and SHAP dependence plots were used to visualize the influence of T and Td on RH modeling. PDPs and ALE revealed a negative correlation between T and RH, and a positive association between Td and RH, while ICE plots revealed localized variability. SHAP plots further captured nonlinear interactions between T and Td. These XAI methods align with scientific knowledge, improving ML model interpretability and accuracy. This framework highlights XAI’s potential to enhance environmental modeling.</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

Explainable AI for gradient boosting-based relative humidity prediction in Fez, Morocco

  • 期刊论文

  • 发表日期：

    2025年5月27日

查看全文
  • Mohamed Chaibi,
  • El Mahjoub Ben Ghoulam,
  • Yassine El Yousfi,
  • Noureddine Khallouk,
  • Jamal Mabrouki,
  • Lhoussaine Tarik,
  • Tarik El-Arrouch,
  • El Arbi Abdellaoui Alaoui

摘要

Relative Humidity (RH) modeling is essential in many fields like meteorology, agriculture, and public health. Machine Learning (ML) techniques have become valuable tools for RH estimation. However, many of these algorithms are hard to interpret and trust due to their black-box nature. This study investigates the potential of eXplainable Artificial Intelligence (XAI) methods to elucidate RH predictions made by a Gradient Boosting (GB) model for Fez, Morocco. RH was estimated via seven input variables: air temperature (T), solar radiation (H), dew temperature (Td), sunshine duration (n), pressure (P), dew temperature (Td), precipitation (Pr), and wind speed (v). T and Td were identified as the most significant predictors using Permutation Feature Importance (PFI), Shapley Additive Global Importance (SAGE), t-statistic-based surrogate modeling, and SHapley Additive exPlanations (SHAP). SHAP methods revealed also that T and Td are the most interacting predictors. The GB model, incorporating only these two features, outperformed (R2 = 0.986, RMSE = 2.166, MAE = 1.621, MAPE = 3.441%) the model utilizing all features (model (R2 = 0.974, RMSE = 2.633, MAE = 2.045, and MAPE = 4.237%). Partial Dependence Plot (PDP), Individual Conditional Expectation (ICE), Accumulated Local Effects (ALE), and SHAP dependence plots were used to visualize the influence of T and Td on RH modeling. PDPs and ALE revealed a negative correlation between T and RH, and a positive association between Td and RH, while ICE plots revealed localized variability. SHAP plots further captured nonlinear interactions between T and Td. These XAI methods align with scientific knowledge, improving ML model interpretability and accuracy. This framework highlights XAI’s potential to enhance environmental modeling.