Stochastic Systems Group
Home Research Group Members Programs  
Demos Calendar Publications Mission Statement Alumni

SSG Seminar Abstract


Performance of Random Forests in the Low False Alarm and Low Missed Detection Regimes

Kush Varshney
SSG, MIT


Different types of errors have different costs in most decision-making problems. In binary classification, there are two types of errors: false alarms and missed detections. We present new analysis of the generalization error of the random forest, a state-of-the-art ensemble classifier, that takes the two types of errors into account. The theoretical analysis is supported by comparison to empirical classification performance, and suggests a design principle for improving performance in either the low false alarm regime or the low missed detection regime.

Joint work with Ryan Prenger, Barry Chen, Tracy Lemmond, and Bill Hanley.



Problems with this site should be emailed to jonesb@mit.edu