Comparison of Neural-Network Learning Algorithms for Time-Series Prediction

Deep learning has received considerable attention in recent years. In this paper, we compare the long short-term memory, a deep learning technique, with other learning techniques such as the back propagation algorithm and the more recently proposed online sequential learning algorithm in the context of time-series prediction. The effectiveness of these learning algorithms is compared using a variety of datasets, univariate and multivariate. We demonstrate that the online sequential learning algorithm is more reliable and provides faster convergence resulting in better prediction performance.