Time Series Prediction Using Coalitions and Self-organizing Maps
Juan C. Burguillo, Juan García‐Rois · Emergence, complexity and computation · 2017
In this chapter we consider the Time Series Prediction problem (TSP), and the use of Self-organizing Maps (SOM) as the basic neural network model to apply. We conduct a topology-based TSP performance analysis, based on extensive numerical simulations, taking into account different complex networks for connecting the neurons within the SOM. We introduce the use of coalitions, and a parameter free version of our Coalitional Algorithm for SOM (CASOM), which adapts the neuron neighborhood to the time series under analysis by means of dynamic coalitions. The results obtained by CASOM are better than the ones provided by SOM in all the topologies and time series considered in this chapter, even when the network structure changes along the training process. Besides, and more remarkable, the number of training epochs, needed by CASOM to stabilize the weights of its neurons, is much lower than SOM. These keywords were added by machine and not by the authors. This process is experimental and the keywords may be updated as the learning algorithm improves.