Learning part-based dictionaries by NMF for crude oil market prediction

Since the crude oil market can make an impact to global economics, it is important to develop some effective approaches to forecast crude oil price and its volatility. In this paper, the goal is to predict the tendency of crude oil future price from ten selected features that potentially affect the crude oil price. Currently, the most popular and robust prediction methods are based on machine learning, such as artificial neural networks, support vector machine, and logistic regression, which are classifiers trained from the training data and used to make predictions for the new data. However, the representations of the data are also crucial to the performance of the classifier training. In this paper, we use non-negative matrix factorization techniques to capture the intrinsic features of the crude oil data, which leads to a part-based dictionary learning problem. Support vector machine (SVM) is trained on the data encoded by the elements from the dictionary in order to predict the tendency of crude oil future price. The experiment shows that the proposed framework is useful for crude oil market prediction.