The Superior Prediction Accuracy of the Random Generalized Linear Model Predictor (RandomGLM)

• Activity Number: 245
• Type: Contributed
• Date/Time: Monday, August 5, 2013 : 2:00 PM to 3:50 PM
• Sponsor: WNAR
• Author(s): Lin Song*+ and Peter Langfelder and Steve Horvath
• Companies: University of California, Los Angeles and Genetics, UCLA and University of California, Los Angeles
• Keywords: RGLM ; machine learning ; ensemble predictor ; generalized linear model

Abstract

Ensemble predictors such as the random forest are known to have superior accuracy but their black-box predictions are difficult to interpret. In contrast, a generalized linear model (GLM) is very interpretable especially when forward feature selection is used. However, forward selection tends to overfit the data and leads to low predictive accuracy. The random generalized linear model (RGLM) combines the advantages of ensemble predictors (high accuracy) with that of forward regression (interpretability). RGLM is a bootstrap aggregated GLM based predictor that incorporates several elements of randomness and instability: random subspace method, optional interaction terms and forward selection. Here we present comprehensive evaluations involving hundreds of genomic data sets and the UCI machine learning benchmark data. RGLM often outperforms alternative methods including random forests, support vector machines, and penalized regression models. RGLM provides variable importance measures that can be used to define a "thinned" ensemble predictor (involving few features) that retains excellent predictive accuracy. These methods are implemented in the R software package randomGLM.