<p>This research explores the use of supervised machine learning algorithms to predict the minimum and maximum temperatures for the next three days in a locality, more precisely in a city using a time-series dataset. Here we have assessed it with 13 weather features spanning 1973–2024 in Kolkata, India. The study used six well-known algorithms— linear regression, decision tree regressor, random forest regressor, bagging regressor, and gradient boosting regressor, finding that the gradient boosting regressor, followed by the multi-layer perceptron, performs the best. Using the past 10 days’ data, these models provided accurate predictions, validated by low error, high <InlineEquation ID="IEq1"> <InlineMediaObject> <ImageObject Color="BlackWhite" FileRef="704_2025_5693_Article_IEq1.gif" Format="GIF" Height="16" Rendition="HTML" Resolution="72" Type="Linedraw" Width="21" /> </InlineMediaObject> <EquationSource Format="TEX">\(R^2\)</EquationSource> </InlineEquation> scores, and Willmott indices. Gradient Boosting tree achieved a Root Mean-Squared Error (RMSE) up to 1.426<InlineEquation ID="IEq2"> <InlineMediaObject> <ImageObject Color="BlackWhite" FileRef="704_2025_5693_Article_IEq2.gif" Format="GIF" Height="14" Rendition="HTML" Resolution="72" Type="Linedraw" Width="23" /> </InlineMediaObject> <EquationSource Format="TEX">\(^{\circ }C\)</EquationSource> </InlineEquation>, Mean Absolute Error (MAE) up to 1.0567<InlineEquation ID="IEq3"> <InlineMediaObject> <ImageObject Color="BlackWhite" FileRef="704_2025_5693_Article_IEq2.gif" Format="GIF" Height="14" Rendition="HTML" Resolution="72" Type="Linedraw" Width="23" /> </InlineMediaObject> <EquationSource Format="TEX">\(^{\circ }C\)</EquationSource> </InlineEquation>, <InlineEquation ID="IEq4"> <InlineMediaObject> <ImageObject Color="BlackWhite" FileRef="704_2025_5693_Article_IEq4.gif" Format="GIF" Height="16" Rendition="HTML" Resolution="72" Type="Linedraw" Width="21" /> </InlineMediaObject> <EquationSource Format="TEX">\(R^{2}\)</EquationSource> </InlineEquation>-score 0.8940, Willmott Index (WI) 0.7924 and a Percent Bias (PBias) of 0.1279 for one-day future prediction. Since time-series data that are interpretable can reveal underlying meteorological patterns, exploring new insights into climate dynamics, explainable AI tools like LIME and SHAP have been studied here which reveal key influencing factors such as average, minimum, and maximum temperature, dewpoint, and wind speed, enhancing model transparency and usability in practical weather forecasting.</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

STAT-X: short-term atmospheric temperature forecasting using machine learning models with explainable-AI

  • Abhiroop Sarkar,
  • Deepsubhra Guha Roy,
  • Piyali Datta

摘要

This research explores the use of supervised machine learning algorithms to predict the minimum and maximum temperatures for the next three days in a locality, more precisely in a city using a time-series dataset. Here we have assessed it with 13 weather features spanning 1973–2024 in Kolkata, India. The study used six well-known algorithms— linear regression, decision tree regressor, random forest regressor, bagging regressor, and gradient boosting regressor, finding that the gradient boosting regressor, followed by the multi-layer perceptron, performs the best. Using the past 10 days’ data, these models provided accurate predictions, validated by low error, high \(R^2\) scores, and Willmott indices. Gradient Boosting tree achieved a Root Mean-Squared Error (RMSE) up to 1.426 \(^{\circ }C\) , Mean Absolute Error (MAE) up to 1.0567 \(^{\circ }C\) , \(R^{2}\) -score 0.8940, Willmott Index (WI) 0.7924 and a Percent Bias (PBias) of 0.1279 for one-day future prediction. Since time-series data that are interpretable can reveal underlying meteorological patterns, exploring new insights into climate dynamics, explainable AI tools like LIME and SHAP have been studied here which reveal key influencing factors such as average, minimum, and maximum temperature, dewpoint, and wind speed, enhancing model transparency and usability in practical weather forecasting.