We have shown that through some simple pre-processing techniques we can use propositional learning systems like M5 to deal with prediction problems. These systems could bring to the field of time series prediction the easy incorporation of background knowledge to improve prediction accuracy. This is a key issue that we hope to elaborate in future work.
Although no extensive variations where tried the results of ML-learned models compare reasonably to other more sophisticated methods.
The results of the experiments carried out are very domain-dependent. They are also influenced by the set of used attributes. Automatic methods for controlling attribute introduction are needed. We outlined a strategy for doing this based on work on feature set selection. With variants of these methods we hope to control the introduction of new attributes.
The analysis of the results of these experiments could lead to the development of a symbolic learner more adapted to prediction tasks.
Acknowledgments
I would like to thank Prof. Pavel Brazdil for is support as well as for reading previous versions of this paper. I would also like to thank Prof. Ross Quinlan for supplying the M5 system.
References
[1]. Box,G., Jenkins,F. (1976) : Time Series Analysis : Forecasting and Control. Oakland, CA. Holden-Day, 1976.
[2]. Breiman,L. , Friedman,J.H., Olshen,R.A. & Stone,C.J. (1984): Classification and Regression Trees, Wadsworth Int. Group, Belmont, California, USA, 1984.
[3]. Caruana,R., Freitag,D. (1994): Greedy Attribute Selection. In Proceedings of 11th IML (International Machine Learning Conference) . Cohen,W. & Hirsh,H. (eds.). Morgan Kaufmann, 1994.
[4]. Dietterich,T., Michalski,R. (1985) : Discovering Patterns in Sequences of Events. In Artificial Intelligence, 25-2. North-Holland, 1985.
[5]. Gershenfeld,N., Weigend,A. (1993) : The Future of Time Series: Learning and Understanding. In Time Series Prediction : Forecasting the Future and Understanding the Past. Weigend,A. & Gershenfeld,N. (eds.). Addison Wesley. 1994.
[6]. John,G., Kohavi,R., Pfleger,K. (1994): Irrelevant Features and the Subset Selection Problem. In Proceedings of 11th IML (International Machine Learning Conference) . Cohen,W. & Hirsh,H. (eds.). Morgan Kaufmann, 1994.
[7]. Judge,G.G, Hill,R.C., Griffiths,W.E, Lutkepohl,H & Lee,T. (1988): Introduction to the Theory and Practice of Econometrics (2nd edition). John Wiley & Sons, 1988.
[8]. Karalic, A.. (1991): The Bayesian Approach to Tree-Structured Regression. In Proceedings of ITI-91. Cavtat, Croatia, 1991.
[9]. Karalic, A..(1992): Employing Linear Regression in Regression Tree Leaves. In Proceedings of ECAI-92. Wiley & Sons, 1992.
[10]. Laird,P., Saul,R. (1994) : Discrete Sequence Prediction and Its Applications. In Machine Learning, 15-1 (p. 43-68). Kluwer Academic Publishers, 1994.
[11]. Meyer,L. Packard,N. (1992) : Local Forecasting of High-Dimensional Chaotic Dynamics. In Nonlinear Modelling and Forecasting. Casdagli,M. & Eubank,S. (eds.). Addison-Wesley, 1992.
[12]. Moore,A., Lee,M. (1994): Efficient Algorithms for Minimizing Cross Validation Error. In Proceedings of 11th IML (International Machine Learning Conference) . Cohen,W. & Hirsh,H. (eds.). Morgan Kaufmann, 1994.
[13]. Mozer,M. (1993) : Neural Net Architectures for Temporal Sequence Processing. In Time Series Prediction : Forecasting the Future and Understanding the Past. Weigend,A. & Gershenfeld,N. (eds.). Addison Wesley. 1994.
[14]. Quinlan, J.R. (1992): Learning with Continuos Classes. In Proceedings of the 5th Australian Joint Conference on Artificial Intelligence. Singapore: World Scientific, 1992.
[15]. Quinlan, J.R. (1993): Combining Instance-Based and Model-Based Learning. In Proceedings of 10th IML (International Machine Learning Conference). Utgoff,P.(ed.). Morgan Kaufmann. 1993.
[16]. Schwarze,J., Weckerle,J. (1982) : Prognoseverfahren im Vergleich. Braunschweig, 1982.
[17]. Weigend,A., Gershenfeld,N. (1994) : Time Series Prediction : Forecasting the Future and Understanding the Past. Addison Wesley. 1994.
[18]. Weigend,A., Huberman,B., Rumelhart,D. (1992) : Predicting Sunspots and Exchange Rates with Connectionist Networks. In Nonlinear Modelling and Forecasting. Casdagli,M. & Eubank,S. (eds.). Addison-Wesley, 1992.