Prediction of rainfall and ore productions (ore and wastes) using multivariate time-series approach, Enyigba open pit mine Ebonyi state, Nigeria

This article aimed at predicting rainfall and ore production of Eyigba open pit Mine Ebonyi State using a Multivariate Time-Series Approach. Rainfall and ore (ore, waste) production data collections, review of machine learning (ML) techniques for  rainfall and ore prediction and integrating of various techniques are few objectives of this article. The datasets are trained, fit into models, propose a model combining Seasonal and Trend Decomposition (STL), Synthetic Minority Over-Sampling Techniques (SMOTE), Random Forest (RF) and Long Term Short Memory (LSTM) in order to optimize the proposed architecture combining Seasonal Trend Decompsotion (STL), Synthetic Minority Over-sampling Technique (SMOTE), Random Forest (RF), and Long Term Short Memory (LSTM), formulate mathematical models for algorithm integration, evaluate performance of the models and their implications for mining operations. The pipeline integrates STL for preprocessing, SMOTE for data augmentation, RF for feature engineering, and LSTM networks for prediction. STL decomposes time series to mitigate seasonality, SMOTE augments underrepresented low-production events, RF selects optimal lags, and LSTM captures temporal dependencies. Performance evaluation of the multivariate model achieves an RMSE of 1,850 tons and MAPE of 8.2% for ore production, surpassing standalone RF(RMSE: 2,450 tons, MAPE: 11.5%) and univariate LSTM (RMSE: 2,120 tons, MAPE: 9.8%). These results demonstrate the framework’s ability to model rainfall-ore interactions, enabling proactive operational strategies. The performance of the models compared and concluded that the integrated LSTM outperformed other models. The findings confirmed that rainfall significantly influences ore production rates due to its impact on equipment efficiency, pit floor stability, and haulage operations.