Research on PM2.5 prediction optimization based on meteorological grouping by multiple linear regression

Accurate prediction of PM2.5 concentration is of great importance for environmental management, public health protection, and sustainable urban development. Meteorological factors such as temperature, humidity, wind speed, and atmospheric pressure play a decisive role in the dispersion and accumulation of pollutants, yet conventional multiple linear regression (MLR) models often fail to account for their temporal heterogeneity and nonlinear interactions. To address this limitation, this study proposes an improved PM2.5 prediction framework that integrates meteorological grouping with MLR. The MiniBatchKMeans algorithm is adopted to cluster meteorological conditions into homogeneous groups, and then separate regression models are constructed for each group. Experimental evaluation is conducted using the Beijing PM2.5 dataset (PRSA_data), covering five years of hourly measurements across multiple seasons and varying pollution episodes. Results show that the proposed grouped regression model outperforms the traditional MLR model, with the coefficient of determination (R²) increasing from 0.4162 to 0.5231, and error metrics (MSE, RMSE, MAE) consistently reduced. Furthermore, residual analysis indicates that the grouped model exhibits more concentrated and stable error distribution compared with the baseline method. These findings confirm that incorporating meteorological grouping into regression not only enhances predictive accuracy but also improves the robustness of PM2.5 forecasting, providing a practical, interpretable, and computationally efficient tool for air quality prediction and early-warning systems.