You are on page 1of 15

MVA analysis of VBS and

ggZZ processes
Setup
• Samples used:

VBS: PolZZjj_LL_0PU.root, PolZZjj_LT_0PU.root, PolZZjj_TT_0PU.root (Pythia)


PolZZjj_LL_200PU.root, PolZZjj_LT_200PU.root, PolZZjj_TT_200PU.root (Pythia)

ggZZ: ggZZ_0PU.root
ggZZ_200PU.root

• Cuts:

Gen level: no cuts (can be implemented if needed)


Reco level: pTe >5 GeV, |ηe|< 4, pTμ >5 GeV, |ηe|< 2.8
MVA setup
• TMVA method: BDT
• Training sample:

signal: VBS signal (ZLZL)


background: only ZTZT events from VBS

Half of sample used for training, half for testing

• MVA application:

signal: VBS signal (ZLZL)


background: VBS background (ZTZT + ZLZT) and ggZZ
MVA setup
• Variables used for signal/background discrimination:

• ηZ
• pTZ
• cosθ*l-
0 PU, training

Training variables
0 PU, training
0 PU, training

WP
For working point cut on BDT output is taken to be 0
0 PU, training
0 PU, application
• As a result of previously defined cut, efficiencies are next:

eff on VBS signal: 44% (3524/7951 events)


eff on VBS background: 14% (18805/136357 events)
eff on ggZZ background: 39% (16778/42627 events)
eff on VBS sig + VBS bkg + ggZZ bkg: 21% (39107/186935 events)
200 PU, training

Training variables
200 PU, training
200 PU, training

WP
For working point cut on BDT output is taken to be 0
200 PU, training
200 PU, application
• As a result of previously defined cut, efficiencies are next:

eff on VBS signal: 44% (3524/7951 events)


eff on VBS background: 14% (18805/136357 events)
eff on ggZZ background: 39% (16778/42627 events)
eff on VBS sig + VBS bkg + ggZZ bkg: 21% (39107/186935 events)
Conclusion
• BDT seems to give some discrimination of VBS signal from
background (VBS and ggZZ)
• As BDT BDT output for VBS signal and ggZZ background is more
simmilar than BDT output for VBS signal and VBS background, it
seems that ggZZ bacground will be harder to eliminate than VBS
signal

You might also like