Short-term Petroleum Price Prediction Based on LSTM and LightGBM

Abstract

With the recent severe fluctuations in the international petroleum market, accurately predicting the direction of oil prices has enormous importance for multiple industries. This study employ the single-unit LSTM model and the multi-unit LSTM model based on word frequency to predict short-term petroleum prices. Results show that the multi-unit LSTM model provides better prediction performance. This study also utilized the LightGBM prediction model taking seven distinct variables into considerate. The result indicates that the model proposed is effective in forecasting crude oil price trends. In conclusion, LSTM model has better interpretability in the time dimension, while the LightGBM model has higher overall prediction accuracy than LSTM. And this study provides ideas for the construction of other oil price forecasting models.