Effect of Hyperparameter Tuning in Long Short-Term Memory for Crude Oil Price Prediction

Forecasting crude oil prices is significant in finance, energy, and economics due to its profound influence on global markets and socioeconomic balance. Using Long Short-Term Memory (LSTM) neural networks has demonstrated remarkable success in time series forecasting, particularly in accurately predicting crude oil prices. However, LSTM models often rely on the manual tuning of hyperparameters, which can be difficult and time-consuming. This study evaluates LSTM networks to optimize the network architecture by looking at lookback variations and train-test split ranges. This study employs historical data on crude oil prices to explore and identify a suitable lookback number by comparing it with the previous study using the same datasets. The experimental findings demonstrate the superiority of the LSTM network using the number of lookbacks equal to 5. It is also compared with statistical time series methods regarding their predictive accuracy. The results indicate that the LSTM model is a valuable resource for financial industries, particularly in oil and gas.