CMStatistics 2018: Start Registration
View Submission - CMStatistics
B0742
Title: Multi-output conditional inference trees applied to the electricity market: Variable importance analysis Authors:  Ismael Ahrazem Dfuf - Technical University of Madrid (Spain) [presenting]
Abstract: Random forests algorithm has been applied extensively due to its high prediction accuracy, interpretability, ability to deal with high dimensional data and to assess the relevance of highly correlated variables in complex non-linear models. We propose an alternative framework to assess the variable importance in multivariate response scenarios based on the permutation importance method using the conditional inference trees algorithm. To build the solution, a $\phi$-divergence measure from information theory is used. The main goal of divergence measures is to provide a distance between probability distributions, in our case, the observations and predicted values. The solution was tested in simulated examples and also in a real case, where we assessed and ranked the most relevant predictors for price and demand of electricity jointly. The results show that the new method outperforms in most cases the outcomes achieved by the recently proposed variable importance technique, Intervention Prediction Measure.