Title: Modal regression based on random forest
Authors: Lin Cong - University of California, Riverside (United States) [presenting]
Weixin Yao - UC Riverside (United States)
Abstract: Modal regression can complement the mean and quantile regressions, and provide a better central tendency measure and prediction performance when the data is skewed or heavy-tailed. Existing nonparametric modal regression has been studied from a kernel density estimation perspective. We propose to utilize a Random Forest-based quantile regression to estimate the global mode of the conditional distribution by minimizing the difference in the quotient of the conditional quantile estimators. The asymptotic property of the model estimator has been studied with its corresponding convergence rate. The validity of the proposed algorithm has been demonstrated through a set of synthetic data analyses and the performance of the algorithm on benchmark data is also compared with some other modal regression models.