Driven Multivariate Regression - Feature Engineering with Random Forest and XGBoost for Accurate Weather Prediction

Fluctuations in weather conditions present a signif- icant challenge for accurate prediction, particularly when con- sidering multiple interdependent variables such as temperature and precipitation. To address this issue, this study proposes a multivariate regression approach driven by advanced feature engineering techniques. Specifically, we employed Random Forest and XGBoost models, leveraging temporal, interaction, and lag features to enhance the predictive accuracy. The proposed method was evaluated using a real-world weather dataset, where both models demonstrated substantial improvements in prediction per-formance. The Random Forest model achieved a Mean Squared Error (MSE) of 0.059 for temperature and 0.019 for precipitation, with an R-squared (R2) of 0.995 and 0.999, respectively. Similarly, the XGBoost model yielded an MSE of 0.043 for temperature and 1.826 for precipitation, with an R2 of 0.996 and 0.983. These results underscore the effectiveness of the feature engineering process combined with robust multivariate regression models, offering a promising approach for accurate weather prediction.