Using LSTM and ARIMA to Simulate and Predict Limestone Price Variations

Abstract

There have been many improvements and advancements in the application of neural networks in the mining industry. In this study, two advanced deep learning neural networks called recurrent neural network (RNN) and autoregressive integrated moving average (ARIMA) were implemented in the simulation and prediction of limestone price variation. The RNN uses long short-term memory layers (LSTM), dropout regularization, activation functions, mean square error (MSE), and the Adam optimizer to simulate the predictions. The LSTM stores previous data over time and uses it in simulating future prices based on defined parameters and algorithms. The ARIMA model is a statistical method that captures different time series based on the level, trend, and seasonality of the data. The auto ARIMA function searches for the best parameters that fit the model. Different layers and parameters are added to the model to simulate the price prediction. The performance of both network models is remarkable in terms of trend variability and factors affecting limestone price. The ARIMA model has an accuracy of 95.7% while RNN has an accuracy of 91.8%. This shows that the ARIMA model outperforms the RNN model. In addition, the time required to train the ARIMA is than that of the RNN. Predicting limestone prices may help both investors and industries in making economical and technical decisions, for example, when to invest, buy, sell, increase, and decrease production.