The traditional energy time series forecasting methods use accurate input data for prediction and then make single-step or multi-step predictions based on the established regression model. But because of the complexity of financial markets, the traditional forecasting methods are less reliable. Unsupervised anomaly detection techniques operate directly on an unseen dataset, under the assumption that outliers are sparsely present in it. A one-class support vector machine (SVM) is an extension of support vector machines for unlabeled data and can be used for anomaly detection. Even though outliers are accounted for in one-class SVMs, they greatly influence the learned model. Hence, one modification to one-class SVMs is presented here: eta one-class SVMs. For evaluation, eta one support vector machine-based algorithm is compared using four kernels. The results affirmed the proficiency of the RBF kernel, with the eta one-class SVM performing best. In regards to subspace division, the preliminary results showed a decreasing nonlinear trend.

Projects

Currently no objects available

Currently no objects available