A Hybrid Traditional and Machine Learning-Based Stacking-Based Ensemble Forecasting Approach for Coal Price Prediction

Abstract

Accurate coal price forecasts are crucial, as volatility in coal prices significantly impacts company performance and profitability. Traditional time series forecasting methods, such as exponential smoothing, are known for their simplicity and low data requirements. In contrast, machine learning techniques, such as random forest and neural network, offer higher accuracy in predictions. However, very few attempts have been made to combine the simplicity of traditional methods with the accuracy of machine learning techniques. This paper presents a novel stacking-based model that integrates both traditional statistical methods and machine learning techniques to enhance coal price predictions. Using Indonesian coal price data from January 2009 to October 2021, we trained the models on various combinations of predictors to generate new predictions. Our findings demonstrate that our stacking-based model outperforms other models, with RMSE and MAPE values of 6.44 and 5.97%, respectively. These results indicate that the model closely forecasts actual coal prices, capturing 94.03% of the price movements. The main contribution of this study is the application of stacking-based models to coal price forecasting in Indonesia, which has not been previously explored, thus enriching the literature on this topic.