1. Variable Choice:

Jet1: Barrel Jet
Jet2: Endcap Jet

From the plots here: run9dijetPtCorrectionTMVAVariableTest.pdf   run9dijetPtCorrectionTMVAVariableTestV2.pdf

I decide to choose Jet2_Pt, Jet2_detEta, Jet1_Pt as the variables in the following studies.

Data sample used: Barrel-Endcap dijet, to faciliate my study, I only use the dijet events which have JP2 endcap jet.

Randomly divide the data sample into two, one for training, the other for testing.

2. Method choice:
Boost Decision Tree(BDT)

1). BDT with Boosting type: Adaboost
run9dijetPtCorrectionTMVAParameterBDTAda.pdf

2). BDT with Boosting type: AdaboostR2
run9dijetPtCorrectionTMVAParameterBDTAdaR2.pdf

3). BDT with Boosting type: Gradient (User Manual: Gradient may not perform very well for regression problems)
run9dijetPtCorrectionTMVAParameterBDTG_test.pdf

Conclusion:
run9dijetPtCorrectionTMVABDT_Conclusion.pdf