Local modeling optimization for time series prediction

The diÆculty of predicting time series generated by chaotic systems has motivated the development of many time series prediction algorithms. Among these local models have emerged as one of the most accurate methods. A weakness of local models is their sensitivity to the choice of user selected parameters such as the size of the neighborhood, the embedding dimension, and the distance metric. This work describes a new method of optimizing these parameters through standard optimization algorithms to minimize the cross-validation error. As a result, the models are more accurate and the user's responsibility to pick sensitive model parameter values is replaced with the responsibility to pick parameters that control the tradeo of accuracy versus computation.