PRIMER – A Regression-Rule Learning System for Intervention Optimization
We introduce intervention optimization as a new area of exploration for data mining research. Interventions are events designed to impact a corresponding time series. The task is to maximize the impact of such events by training a model on historical data. We propose PRIMER as a new regression-rule learning system for identifying sets of event features that maximize impact. PRIMER is for use when domain experts with knowledge of the intervention can specify a transfer function, or the form of the expected response in the time series. PRIMER’s objective function includes the goodness-of-fit of the average response of covered events to the transfer function. Incorporating domain knowledge in this way makes PRIMER robust to over-fitting on noise or spurious responses. PRIMER is designed to produce interpretable results, improving on the interpretability of even competing regression-rule systems for this task. It also has fewer and more intuitive parameters than competing rule-based systems. Empirically, we show that PRIMER is competitive with state-of-the-art regression techniques in a large-scale event study modeling the impact of insider trading on intra-day stock returns.
KeywordsRegression rules Intervention analysis Rule induction Event response Time series Intervention optimization Rule learning
This work is supported by Chevron USA, Inc. under the joint project Center for Interactive Smart Oilfield Technologies (CiSoft), at the University of Southern California.
We would also like to thank Dr. Frederik Janssen for providing support with the SeCoReg and Dynamic Reduction to Regression algorithms.
- 2.Brookhouse, J., Otero, F.E.B.: Discovering regression rules with ant colony optimization. In: Proceedings of the Companion Publication of the 2015 Annual Conference on Genetic and Evolutionary Computation, GECCO Companion 2015, pp. 1005–1012. ACM, New York (2015)Google Scholar
- 3.Clark, P., Niblett, T.: The CN2 induction algorithm. Mach. Learn. 3(4), 261–283 (1989)Google Scholar
- 13.Janssen, F., Fürnkranz, J.: A re-evaluation of the over-searching phenomenon in inductive rule learning. In: SDM, pp. 329–340. SIAM (2009)Google Scholar
- 14.Janssen, F., Fürnkranz, J.: Separate-and-conquer regression. In: Proceedings of LWA 2010: Lernen, Wissen, Adaptivität, Kassel, Germany, pp. 81–89 (2010)Google Scholar
- 15.Janssen, F., Fürnkranz, J.: Heuristic rule-based regression via dynamic reduction to classification. In: IJCAI Proceedings-International Joint Conference on Artificial Intelligence, vol. 22, p. 1330 (2011)Google Scholar
- 17.Možina, M., Demšar, J., Žabkar, J., Bratko, I.: Why is rule learning optimistic and how to correct it. In: Fürnkranz, J., Scheffer, T., Spiliopoulou, M. (eds.) ECML 2006. LNCS (LNAI), vol. 4212, pp. 330–340. Springer, Heidelberg (2006)Google Scholar
- 18.Quinlan, J., Cameron-Jones, R.: Oversearching and layered search in empirical learning. Breast Cancer 286, 2–7 (1995)Google Scholar