An Efficient Model for Time Series Prediction using Instance Based Learning Technique

Abstract

Multi-step ahead time series forecasting has become an important activity in various fields of science and technology, due to its usefulness in future events management. Long-term or multi-step prediction problem is a challenging area as it predicts several steps ahead into the future -starting from information at current instant. Storing and using specific instances improves the performance of several supervised learning algorithms. Instance-Based learning is a framework and methodology that can be applied to generate time series predictions using specific instances. In this paper we propose a new concept to improve the performance of prediction model. The proposed learning technique implemented here extends the nearest neighbour algorithm to include the concept of pattern matching to identify similar instances thus implementing a nonparametric regression approach. Pattern matching in the context of time-series forecasting implies the process of matching current state of the time series with its past states. Specific instances are chosen using hybrid distance measure which includes correlation measure with that of Euclidean distance measure. The instances chosen are combined using multiple regressions to generate multi-step ahead predictions. Bench mark data set of Mackey-Glass series and real time series of Nord Pool are used to test and validate the proposed technique. Experimental results show remarkable enhancement in the performance of our prediction model.

 

Keywords: Instance Based Learning (IBL), Multistep ahead Prediction, Time series forecasting.