A Novel Approach to Forecasting Crude Oil Price Based on LSTM and Online Learning

Crude Oil, a mixture of petroleum liquid and gases, extracted from the ground, is undoubtedly the fuel that drives the modern civilization. It affects the economic situation, often very directly in many countries. In this paper, we introduce a novel approach to forecasting crude oil price by combining existing machine learning paradigms namely, Online Learning and Long Short-Term Memory (LSTM). In our approach, we combine LSTM with a sliding window approach, so that the model gets updated as it receives new data, and captures the changing trends as soon as a new price is available in a more accurate way. The experiment results, compared with the existing LSTM Model of the same architecture using performance metrics like Mean Absolute Error, as well as Root Mean Squared Error and Directional Accuracy (DA), show that our model, with an RMSE of 1.65, MAE of 1.36 and the Directional Accuracy of 65.52% for the best scenarios dominates in terms of less error rate and more accuracy and achieves better Directional Accuracy.