Estimating the horizon of predictability in time-series predictions using inductive modelling tools

Abstract

This paper deals with the assessment of how far into the future a time series can be safely predicted using inductive modelling and extrapolation techniques. Three different time series are used to demonstrate the viability of the approaches presented in the paper: one time series representing the water demand of the city of Barcelona, another characterizing the water demand of a section of the city of Rotterdam, and a third describing weather data for the city of Tucson. Fuzzy inductive reasoning (FIR) is used to predict future values of these time series on the basis of their own past. FIR predictions come with two different built-in measures of confidence that can be used to obtain a quantitative estimate of how far into the future a time series can be predicted.

Keywords:

• inductive modelling
• time-series prediction
• fuzzy inductive reasoning
• estimation of predictability horizon

Acknowledgements

The authors wish to thank Aigües de Barcelona S.A. and Waterbedrijf Europoort for providing the water demand data files used in this investigation.