Hybrid Climate Forecasting using Ensemble Learning and Trend Detection
Hasan Ahamed Alif, V. Paramasivam, Sugumar Rajendran, Anik Dev Nath, Md Assaduzzaman
Abstract
Understanding local climate change patterns is vital for establishing an effective adaptation plan, especially in climate-sensitive places like Rajshahi, Bangladesh. This study examines monthly temperature data from 1980 to 2024 to forecast future climatic conditions and find long-term trends using hybrid analytics, which integrates strong machine learning algorithms with standard statistical approaches. Sen's Slope estimator, Linear Regression, and the statistical model Mann-Kendall test were applied to do trend analysis during the first phase. These techniques exhibited a substantial growing trend in both maximum and minimum temperatures, which indicates a long-term warming signal. Three supervised models were employed to predict monthly temperatures for 2025–2027: Extreme Gradient Boosting (XGBoost), Random Forest (RF), and Linear Regression (LR). The study employed three performance evaluation matrices for measuring model performance: Mean Absolute Error (MAE), Root Mean Squared Error (RMSE), and R<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup> Score. Random Forest was best at the lowest temperature (R2 = 0.997, RMSE = 0.101°C). On the other hand, XGBoost performs well at the maximum temperature (R2 = 0.994, RMSE = 0.154°C) with exceptional precision. By studying the findings, this research acknowledges that the data-driven strategy at the regional climate trend assessment boosts the forecasting and trend analysis reliability by merging both machine learning and statistical techniques. Deep learning models and integrating new meteorological variables will be vital for future research to enhance prediction accuracy and scalability.