Speaker Biography
Prof Markus Löcher is a professor of mathematics and statistics at the Berlin School of Economics and Law (HWR Berlin) since 2011. His research includes machine learning, spatial statistics, data visualization and sequential learning. Prior to joining HWR Berlin he worked as principal and lead scientist at various data analytics companies in the United States. He holds a PhD in physics and a master degree in statistics.
Talk Abstract
Random forests and boosting algorithms are becoming increasingly popular in many scientific fields because they can cope with “small n large p” problems, complex interactions and even highly correlated predictor variables. The predictive power of covariates derives from two essentially different feature importance score in random forests. It has been shown that these variable importance measures show a bias towards correlated predictor variables. We demonstrate the fundamental dilemma of variable importance measures as well as their appeal and wide spread use in practical data science applications. We address recent criticism of the reliability of these scores by residualizing and deriving analogous procedures to the F-test.