Using LSTM and ARIMA to Simulate and Predict Limestone Price Variations.

Area of Science:

• Data Science
• Artificial Intelligence
• Econometrics

Background:

• Neural networks and deep learning have seen significant advancements in the mining sector.
• Accurate prediction of commodity prices like limestone is crucial for economic and technical decision-making.

Purpose of the Study:

• To implement and compare the performance of two advanced models, Recurrent Neural Network (RNN) and Autoregressive Integrated Moving Average (ARIMA), for limestone price prediction.
• To evaluate the effectiveness of these models in simulating price variations and identifying influencing factors.

Main Methods:

• Recurrent Neural Network (RNN) utilizing Long Short-Term Memory (LSTM) layers, dropout regularization, activation functions, Mean Square Error (MSE), and the Adam optimizer.
• Autoregressive Integrated Moving Average (ARIMA) as a statistical time series model, employing an auto ARIMA function to identify optimal parameters.
• Both models were configured with various layers and parameters for price simulation and prediction.

Main Results:

• Both ARIMA and RNN models demonstrated remarkable performance in simulating trend variability and factors influencing limestone prices.
• The ARIMA model achieved a higher accuracy of 95.7% compared to the RNN model's 91.8%.
• The ARIMA model required less training time than the RNN model.

Conclusions:

• The ARIMA model is more effective than the RNN model for limestone price prediction, offering superior accuracy and efficiency.
• Accurate limestone price forecasting can significantly aid investors and industries in making informed decisions regarding investment, production, and market strategies.