Predicting the future with the appropriate embedding dimension and time lag

Prediction is a typical example of a generalization problem. The goal of prediction is to accurately forecast the short-term evolution of the system based on past information. Neural network and fuzzy logic techniques are used because they both have good generalization capabilities. The embedding dimension (number of inputs) and the time lag selection problem is treated in this paper. It is proposed that the selection of the appropriate embedding dimension and time lag for the input/output space construction plays an important role in the performance of the above networks. It is shown that the “traditionally accepted” choices for the embedding dimension and time lag are not optimal. The proposed method offers an improvement over the traditionally accepted parameter choices. Different analytical techniques for the determination of these parameters are used, and the results are evaluated