Using Boosted Decision Trees to Separate Signal and Background in B to XsGamma Decays
The measurement of the branching fraction of the flavor changing neutral current B {yields} X{sub s}{gamma} transition can be used to expose physics outside the Standard Model. In order to make a precise measurement of this inclusive branching fraction, it is necessary to be able to effectively separate signal and background in the data. In order to achieve better separation, an algorithm based on Boosted Decision Trees (BDTs) is implemented. Using Monte Carlo simulated events, ''forests'' of trees were trained and tested with different sets of parameters. This parameter space was studied with the goal of maximizing the figure of merit, Q, the measure of separation quality used in this analysis. It is found that the use of 1000 trees, with 100 values tested for each variable at each node, and 50 events required for a node to continue separating give the highest figure of merit, Q = 18.37.
- Research Organization:
- Stanford Linear Accelerator Center (SLAC)
- Sponsoring Organization:
- USDOE
- DOE Contract Number:
- AC02-76SF00515
- OSTI ID:
- 892609
- Report Number(s):
- SLAC-TN-06-015
- Country of Publication:
- United States
- Language:
- English
Similar Records
Fast inference of Boosted Decision Trees in FPGAs for particle physics
LeptonJet Categorization with Decision Tree