A Wavelet PM2.5 Prediction System Using Optimized Kernel Extreme Learning with Boruta-XGBoost Feature Selection

The fine particulate matter (PM2.5) concentration has been a vital source of info and an essential indicator for measuring studying the other air pollutants. It is crucial to realize more accurate predictions PM2.5 establish high-accuracy prediction model due their social impacts cross-field applications in geospatial engineering. To further boost accuracy results, this paper proposes new wavelet system (called WD-OSMSSA-KELM model) based on new, improved variant salp swarm algorithm (OSMSSA), kernel extreme learning machine (KELM), decomposition, Boruta-XGBoost (B-XGB) feature selection. First, we applied B-XGB selection best features predicting hourly concentrations. Then, decomposition (WD) reach multi-scale results single-branch reconstruction concentrations mitigate error produced by time series data. In next stage, optimized parameters KELM under each reconstructed component. An version SSA proposed higher performance basic optimizer avoid local stagnation problems. work, propose operators oppositional-based simplex-based search core problems conventional SSA. addition, utilized time-varying parameter instead main exploration trends SSA, using random leaders guide towards regions space conditional structure. After optimizing model, was predict concentrations, different metrics were evaluate model’s accuracy. evaluated database, six pollutants, meteorological collected from Beijing Municipal Environmental Monitoring Center. experimental show that WD-OLMSSA-KELM can with superior (R: 0.995, RMSE: 11.906, MdAE: 2.424, MAPE: 9.768, KGE: 0.963, R2: 0.990) compared WD-CatBoost, WD-LightGBM, WD-Xgboost, WD-Ridge methods.