Determined prediction of nonlinear time series via emotional temporal difference learning

Bounded rationality and satisfying models, rather than optimization techniques, have shown good performance in decision-making, for which the emotional temporal learning algorithm is an example. This is based on simulating human emotions via reinforcement agents. A new approach towards deterministic prediction problems, derived from a recently developed model of emotional temporal learning in human brain is introduced in this paper. The proposed algorithm inherently emphasizes on learning to predict future peaks, and performs remarkably accurate predictions among important regions, features, or objectives. Stock market and foreseeing solar activity are excellent examples of using this methodology, and in fact were the motivation to introduce purposeful prediction via multi-objective learning algorithms in this research. With two examples, we show the characteristics of the suggested algorithm and its usefulness to space weather warning and alert system. The successful application of our proposed model indicates the significance of structural brain modeling beyond traditional neural networks, as well as the importance of biological motivation in choosing a suitable model for any given application.