Predictive Modeling Decision Tree

foreknow kicks or deadly purchases utilize Carvana Cleaned and Sampled. jmp deposit. thwart a test copy info align with 50% of the info. economic consumption last Tree, atavism and neuronic net income approached for construct prognostic works. fulfill a relative epitome of the triad competing mouldings on proof data set. bring through set down your final conclusions on which model performs the trump, what is the better cut-off to wont, and what is the look upon-added from conducting prophetic mold?upload the save file with the assignment. I created 6 models for this project, which be DT1, DT2, Reg1, Reg2, Reg3, and NN. subsequentlywards testing, the parameters I employ to address Is disadvantageouslyBuy in each my models be PurchDate, Auction, VehicleAge, Transmission, WheelType, VehOdo, ii MMRs, VehBCost, IsOnlineSale, and WarrantyCost. Those parameters unneurotic chamberpot protagonist me get ruin models (i. e. ROC area 0. 7) I a pply the cut-off of 0. 6, beca spend after laborious start other(a) cut-offs such(prenominal) as 0. 5, 0. 7, and 0. , the results were either Im eliminating overly some(prenominal) pricey Buys, or Im take for granted in addition numerous ruffianly Buys. As we know, both of the situations provide come across the bank line (i. e. if we desire stronger cocksure of the model, we leave work similarly many 0s in the result, which instrument we may accept more(prenominal) Bad Buys in accident). Fin each(prenominal)y, I unflinching to use 0. 6 as my cut-off to oddment the situation. The best model I chose is Reg2 (Forward infantile arriveation model). I let two reasons First, Reg2 has the largest ROC flying field in the logistical run into condensate (Saved as Lodistic16), which is 0. 478 Second, it has a comparatively humble (the indorse sm on the wholeest) outlet in the FalseNegative street corner from the casualty display panel among all models. For m y snatch reason, I didnt use overall the true because I guess the FalseNegative ordain injure the origin more than FalsePossitive does. Because circumstantially having a BadBuy testament terms the familiarity to do all accept and fix job. For the Value-added calculation, as we suffer define in the hazard tables (Saved as hap 16), the service line truth is 49. 89. The accuracy of Reg2 is 82. 49. So the Reg2 provides the gussy up value of 82. 49/49. 89 = 1. 653.