Deep Learning Based Prediction of Commodity Prices Using LSTM

• Published in: 2023 4th International Conference on Smart Electronics and Communication (ICOSEC) pp. 1708 - 1715
• Main Authors: Deepa, P.B., Daisy, Josephine
• Format: Conference Proceeding
• Language: English
• Published: IEEE 20.09.2023
• Subjects: Adaptation models; Computational modeling; Convolutional neural network (CNN); Deep learning; Finance; Long short-term memory (LSTM); Predictive models; Solid modeling; Stock market; Transfer learning; Uncertainty
• DOI: 10.1109/ICOSEC58147.2023.10276011

Commodity price prediction using deep learning, specifically LSTM models, is a challenging task due to limited data, price volatility, and the complexity of capturing temporal dependencies. In recent years, various techniques have been explored to address these challenges, such as transfer learning, model interpretability, handling seasonality, and ensemble learning. However, existing systems still face issues related to model interpretability, uncertainty quantification, and adaptability to evolving market conditions. This article aims to analyze one of the machine learning methods and what applications it can have in the financial world. Specifically, it aims to explore a technique called deep learning. To do this, some methods are analyzed, which are most relevant and most used in the industry, which are found within the deep learning method. Once the selected methods are analyzed, it is intended to implement them in a programming environment, in this case Python is chosen, and then a practical case is developed and evaluated. The case in question is the forecasting of commodity prices, with particular focus on the price of Brent. Different technical analysis indicators will be used in this forecasting model, knowing that there is no solid statistical basis in their use, but in practice most traders use them and, in the absence of shocks, on a macroeconomic basis, the results obtained when using them are, in general, quite adequate. Finally, the results of the predictions obtained are compared under different models to calculate which model gives us the greatest advantage.