analysis of h ww l n l n based on boosted decision trees
DESCRIPTION
Analysis of H WW l n l n Based on Boosted Decision Trees. Hai-Jun Yang University of Michigan (with T.S. Dai, X.F. Li, B. Zhou) ATLAS Higgs Meeting September 10, 2007. Outline. Physics Motivation MC samples for H WW analysis Introduction of Boosted Decision Trees (BDT) - PowerPoint PPT PresentationTRANSCRIPT
![Page 1: Analysis of H WW l n l n Based on Boosted Decision Trees](https://reader035.vdocuments.mx/reader035/viewer/2022062805/56814e32550346895dbb98c5/html5/thumbnails/1.jpg)
Analysis of H WW ll Based on Boosted Decision Trees
Hai-Jun Yang
University of Michigan(with T.S. Dai, X.F. Li, B. Zhou)
ATLAS Higgs Meeting
September 10, 2007
![Page 2: Analysis of H WW l n l n Based on Boosted Decision Trees](https://reader035.vdocuments.mx/reader035/viewer/2022062805/56814e32550346895dbb98c5/html5/thumbnails/2.jpg)
Outline
• Physics Motivation
• MC samples for H WW analysis
• Introduction of Boosted Decision Trees (BDT)
• H WW analysis based on BDT
• Sensitivity of H WW ll• Summary
![Page 3: Analysis of H WW l n l n Based on Boosted Decision Trees](https://reader035.vdocuments.mx/reader035/viewer/2022062805/56814e32550346895dbb98c5/html5/thumbnails/3.jpg)
The Standard Model
Why particles have mass?Higgs mechanism ?
If SM Higgs boson exists, the LHC will be able to detect it.
![Page 4: Analysis of H WW l n l n Based on Boosted Decision Trees](https://reader035.vdocuments.mx/reader035/viewer/2022062805/56814e32550346895dbb98c5/html5/thumbnails/4.jpg)
Higgs Production at LHC
Having available four production mechanismsis a key for measurements of Higgs parameters
![Page 5: Analysis of H WW l n l n Based on Boosted Decision Trees](https://reader035.vdocuments.mx/reader035/viewer/2022062805/56814e32550346895dbb98c5/html5/thumbnails/5.jpg)
BR and Discovery Channelsm(H) > 2 mZ
H ZZ 4 qqH ZZ * qqH ZZ jj *
qqH WW jj *
* for mH > 300
GeV
forward jet tag
Low mass region: m(H) < 2 mZ
H H bbH H ZZ* 4H WW* or jj
-
![Page 6: Analysis of H WW l n l n Based on Boosted Decision Trees](https://reader035.vdocuments.mx/reader035/viewer/2022062805/56814e32550346895dbb98c5/html5/thumbnails/6.jpg)
H WW ll• Signal H WW using CSC V12• All background MC using CSC V11• gg WW MC are not available for this study• Full ATLAS detector simulation
• H WW eee• Higgs production based on gg fusion & VBF• Higgs masses:
– 140, 150, 160, 165, 170, 180 GeV
![Page 7: Analysis of H WW l n l n Based on Boosted Decision Trees](https://reader035.vdocuments.mx/reader035/viewer/2022062805/56814e32550346895dbb98c5/html5/thumbnails/7.jpg)
Pre-selection Efficiency
• At least one lepton pair (ee, , e) with Pt>10 GeV • Missing Et > 15 GeV• |Mee – Mz| > 10 GeV, |M – Mz| > 15 GeV to suppress background from Z ee,
![Page 8: Analysis of H WW l n l n Based on Boosted Decision Trees](https://reader035.vdocuments.mx/reader035/viewer/2022062805/56814e32550346895dbb98c5/html5/thumbnails/8.jpg)
MC for HWWeenalysis
Higgs Signal
SM Background
![Page 9: Analysis of H WW l n l n Based on Boosted Decision Trees](https://reader035.vdocuments.mx/reader035/viewer/2022062805/56814e32550346895dbb98c5/html5/thumbnails/9.jpg)
Event Weight for 1fb-1
dtN
NBrKN
MC
precuttest L
1 fb-1
11 beam-days @ 1033
preselect
test
N
NWeight # events in 1 fb-
1
![Page 10: Analysis of H WW l n l n Based on Boosted Decision Trees](https://reader035.vdocuments.mx/reader035/viewer/2022062805/56814e32550346895dbb98c5/html5/thumbnails/10.jpg)
MC for HWWnalysis
Higgs Signal
SM Background
![Page 11: Analysis of H WW l n l n Based on Boosted Decision Trees](https://reader035.vdocuments.mx/reader035/viewer/2022062805/56814e32550346895dbb98c5/html5/thumbnails/11.jpg)
MC for HWWenalysis
Higgs Signal
SM Background
![Page 12: Analysis of H WW l n l n Based on Boosted Decision Trees](https://reader035.vdocuments.mx/reader035/viewer/2022062805/56814e32550346895dbb98c5/html5/thumbnails/12.jpg)
BDT Training Variables
![Page 13: Analysis of H WW l n l n Based on Boosted Decision Trees](https://reader035.vdocuments.mx/reader035/viewer/2022062805/56814e32550346895dbb98c5/html5/thumbnails/13.jpg)
Signal vs Background
X
![Page 14: Analysis of H WW l n l n Based on Boosted Decision Trees](https://reader035.vdocuments.mx/reader035/viewer/2022062805/56814e32550346895dbb98c5/html5/thumbnails/14.jpg)
Boosted Decision Trees (BDT) How to build a decision tree ?For each node, try to find the best variable and splitting point which gives the best separation based on Gini index.Gini_node = Weight_total*P*(1-P), P is weighted purityCriterion = Gini_father – Gini_left_son – Gini_right_sonVariable is selected as splitter by maximizing the criterion.
How to boost the decision trees?Weights of misclassified events in current tree are increased, thenext tree is built using the same events but with new weights.Typically, one may build few hundred to thousand trees.
How to calculate the event score ?For a given event, if it lands on the signal leaf in one tree, it is
given a score of 1, otherwise, -1. The sum (probably weighted) of scores from all trees is the final score of the event.
Ref: B.P. Roe, H.J. Yang, J. Zhu, Y. Liu, I. Stancu, G. McGregor, ”Ref: B.P. Roe, H.J. Yang, J. Zhu, Y. Liu, I. Stancu, G. McGregor, ”Boosted decision trees as an alternative to Boosted decision trees as an alternative to artificial neural networks for particle identificationartificial neural networks for particle identification”, physics/0408124, NIM A543 (2005) 577-584.”, physics/0408124, NIM A543 (2005) 577-584.
Sum of 1000 trees
![Page 15: Analysis of H WW l n l n Based on Boosted Decision Trees](https://reader035.vdocuments.mx/reader035/viewer/2022062805/56814e32550346895dbb98c5/html5/thumbnails/15.jpg)
Weak Powerful Classifier
Boosted decision trees focus on the misclassified events which usually have high weights after hundreds of tree iterations. An individual tree has a very weak discriminating power; the weighted misclassified event rate errm is about 0.4-0.45.
The advantage of using boosted decision trees is that it combines many decision trees, “weak” classifiers, to make a powerful classifier. The performance of boosted decision trees is stable after a few hundred tree iterations.
Ref1: H.J.Yang, B.P. Roe, J. Zhu, “Ref1: H.J.Yang, B.P. Roe, J. Zhu, “Studies of Boosted Decision Trees for MiniBooNE Particle IdentificationStudies of Boosted Decision Trees for MiniBooNE Particle Identification”, ”, physics/0508045, Nucl. Instum. & Meth. A 555(2005) 370-385.physics/0508045, Nucl. Instum. & Meth. A 555(2005) 370-385.Ref2: H.J. Yang, B. P. Roe, J. Zhu, Ref2: H.J. Yang, B. P. Roe, J. Zhu, " Studies of Stability and Robustness for Artificial Neural Networks " Studies of Stability and Robustness for Artificial Neural Networks and Boosted Decision Trees "and Boosted Decision Trees ", physics/0610276, Nucl. Instrum. & Meth. A574 (2007) 342-349., physics/0610276, Nucl. Instrum. & Meth. A574 (2007) 342-349.
![Page 16: Analysis of H WW l n l n Based on Boosted Decision Trees](https://reader035.vdocuments.mx/reader035/viewer/2022062805/56814e32550346895dbb98c5/html5/thumbnails/16.jpg)
BDT Training with Event Reweighting
• In the original BDT training program, all training events are set to have same weights in the beginning (the first tree). It works fine if all MC processes are produced based on their production rates.
• Our MCs are produced separately, the event weights vary from various backgrounds. e.g. assuming 1 fb-1 wt (WW) = 0.029, wt (ttbar) = 0.61, wt(DY) = 12.1
• If we treat all training events with different weights equally using “standard” training algorithm, ANN/BDT tend to pay more attention to events with lower weights (high stat.) and introduce training prejudice.
• Ref: http://arxiv.org/abs/0708.3635, Hai-Jun Yang, Tiesheng Dai, Alan Wilson, Zhengguo Zhao, Bing Zhou, ”A Multivariate Training Technique with Event Reweighting”
![Page 17: Analysis of H WW l n l n Based on Boosted Decision Trees](https://reader035.vdocuments.mx/reader035/viewer/2022062805/56814e32550346895dbb98c5/html5/thumbnails/17.jpg)
Higgs Mass = 140 GeV
![Page 18: Analysis of H WW l n l n Based on Boosted Decision Trees](https://reader035.vdocuments.mx/reader035/viewer/2022062805/56814e32550346895dbb98c5/html5/thumbnails/18.jpg)
Higgs Mass = 150 GeV
![Page 19: Analysis of H WW l n l n Based on Boosted Decision Trees](https://reader035.vdocuments.mx/reader035/viewer/2022062805/56814e32550346895dbb98c5/html5/thumbnails/19.jpg)
Higgs Mass = 160 GeV
![Page 20: Analysis of H WW l n l n Based on Boosted Decision Trees](https://reader035.vdocuments.mx/reader035/viewer/2022062805/56814e32550346895dbb98c5/html5/thumbnails/20.jpg)
Higgs Mass = 165 GeV
![Page 21: Analysis of H WW l n l n Based on Boosted Decision Trees](https://reader035.vdocuments.mx/reader035/viewer/2022062805/56814e32550346895dbb98c5/html5/thumbnails/21.jpg)
Higgs Mass = 170 GeV
![Page 22: Analysis of H WW l n l n Based on Boosted Decision Trees](https://reader035.vdocuments.mx/reader035/viewer/2022062805/56814e32550346895dbb98c5/html5/thumbnails/22.jpg)
Higgs Mass = 180 GeV
![Page 23: Analysis of H WW l n l n Based on Boosted Decision Trees](https://reader035.vdocuments.mx/reader035/viewer/2022062805/56814e32550346895dbb98c5/html5/thumbnails/23.jpg)
Confidence Level Calculation
Log-likelihood ratio test-statistics using 10 BDT bins
MC experiments are based on Poisson statistics
CLb represents C.L. to exclude “background only” hypothesis
![Page 24: Analysis of H WW l n l n Based on Boosted Decision Trees](https://reader035.vdocuments.mx/reader035/viewer/2022062805/56814e32550346895dbb98c5/html5/thumbnails/24.jpg)
Sensitivity of H WW ll
![Page 25: Analysis of H WW l n l n Based on Boosted Decision Trees](https://reader035.vdocuments.mx/reader035/viewer/2022062805/56814e32550346895dbb98c5/html5/thumbnails/25.jpg)
Summary
• H WW ll analysis based on BDT
• For 165 GeV Higgs, 5 sigma discovery can be achieved using about 1.1 fb-1 data.
• Major backgrounds for H WW searches come from ttbar(40%-50%), WW(10%-20%), Drell-Yan(20%-30%), W l (15%-20%)
• Additional BDTs trained with Higgs signal against major backgrounds may help to further suppress background and improve sensitivity.
![Page 26: Analysis of H WW l n l n Based on Boosted Decision Trees](https://reader035.vdocuments.mx/reader035/viewer/2022062805/56814e32550346895dbb98c5/html5/thumbnails/26.jpg)
Backup Slides
![Page 27: Analysis of H WW l n l n Based on Boosted Decision Trees](https://reader035.vdocuments.mx/reader035/viewer/2022062805/56814e32550346895dbb98c5/html5/thumbnails/27.jpg)
BDT Free Software
• http://gallatin.physics.lsa.umich.edu/~hyang/boosting.tar.gz
• TMVA toolkit, CERN Root http://tmva.sourceforge.net/http://root.cern.ch/root/html/src/TMVA__MethodBDT.cxx.html
![Page 28: Analysis of H WW l n l n Based on Boosted Decision Trees](https://reader035.vdocuments.mx/reader035/viewer/2022062805/56814e32550346895dbb98c5/html5/thumbnails/28.jpg)
Properties of the Higgs Boson
The decay properties of the Higgs boson are fixed,if the mass is known:
W+, Z, t, b, c, +,..........., g
W-, Z, t, b, c, - ,.........., g
H
Higgs Boson:
• it couples to particles proportional to their masses
• decays preferentially in the heaviest particles kinematically allowed