EOF-Enhanced XGBoost for Air Quality Prediction in a Data-Constrained Mining Environment: Evidence from Gujeni Ion-ore Mining Area, Kagarko Local Government Area, Kaduna State- Nigeria

Authors

  • David Joseph Ariko Federal University of Applied Sciences, Kachia Author https://orcid.org/0009-0005-3956-5214
  • Samson Isaac Federal University of Applied Sciences, Kachia Author
  • Thomas Barna Lass Federal University of Applied Sciences, Kachia Author

DOI:

https://doi.org/10.47514/kjg.2026.08.01.075

Keywords:

XGBoost; empirical orthogonal functions; GeoAI; air-quality prediction; explainable AI; AQI; meteorology; mining; Gujeni; Nigeria

Abstract

Predictive air-quality modelling in mining environments is challenged by nonlinear meteorological interactions and correlated predictors. This study developed and evaluated an Empirical Orthogonal Function (EOF)-enhanced Extreme Gradient Boosting (XGBoost) framework for predicting a screening Air Quality Index (AQI) in the Gujeni mining environment, Kaduna State, Nigeria. Monthly pollutant and meteorological data covering 1986–2025 were obtained from MERRA-2 and ERA5-Land. Four climatic predictors; temperature, rainfall, relative humidity and wind speed were standardised and transformed using EOF analysis, retaining components explaining at least 90% of cumulative variance. XGBoost models were evaluated using training, testing and validation subsets, with grid-search optimisation applied to model hyperparameters. The default model achieved testing R² = 0.65 and RMSE = 21.96, while the tuned model improved testing R² to 0.81 and reduced RMSE to 14.82. The EOF-enhanced model achieved validation R² = 0.89, RMSE = 10.31, MAE = 7.42, NSE = 0.89, KGE = 0.92 and Pearson r = 0.95. Gain-based feature importance ranked relative humidity first (38.8%), followed by rainfall (27.4%), wind speed (20.8%) and temperature (13.0%). The results demonstrate that predictor optimisation and dimensionality reduction can materially improve AQI prediction in the study dataset. However, the reported importance values are gain-based rather than SHAP values, and the model has not been independently validated against ground observations or a strictly chronological out-of-sample design. The framework should therefore be regarded as a screening and decision-support approach pending independent validation.

Downloads

Published

2026-11-19

How to Cite

Ariko, D. J., Isaac, S., & Lass, T. B. (2026). EOF-Enhanced XGBoost for Air Quality Prediction in a Data-Constrained Mining Environment: Evidence from Gujeni Ion-ore Mining Area, Kagarko Local Government Area, Kaduna State- Nigeria. Kaduna Journal of Geography, 8(1), 803-812. https://doi.org/10.47514/kjg.2026.08.01.075