B1757
Title: Boosting diversity in regression ensembles
Authors: Jean-Michel Poggi - University Paris-Saclay Orsay (France) [presenting]
Jairo Cugliari - Université Lumière Lyon 2 (France)
Yannig Goude - EDF (France)
Mathias Bourel - University de la Republica (Uruguay)
Abstract: The practical interest in using ensemble methods has been highlighted in several works. Aggregation estimation, as well as sequential prediction, provide natural frameworks for studying ensemble methods and for adapting such strategies to time series data. Sequential prediction focuses on how to combine by weighting a given set of individual experts while aggregation is mainly interested in how to generate individual experts to improve prediction performance. We look for enhancing these (possibly online) mixture methods by using the concept of diversity. We propose an algorithm to enrich the set of original individual predictors using a gradient boosting-based method by incorporating a diversity term to guide the gradient boosting iterations. The idea is to progressively generate experts by boosting diversity. Then, we establish a convergence result ensuring that the associated optimization strategy converges to a global optimum. Finally, we show by means of numerical experiments the appropriateness of our procedure using simulated data and real-world electricity demand datasets.
