Cox Proportional Hazard with Multivariate Adaptive Regression Splines to Analyze the Product Sales Time in E-Commerce

Abstract

Cox Proportional Hazard (Cox PH) model is a survival analysis method to perform model of relationship between independent variable and dependent variable which shown by time until an event occurs. This method compute residuals, martingale or deviance, which can used to diagnostic the lack of fit of a model and PH assumption. The alternative method if these not satisfied is Multivariate Adaptive Regression Splines(MARS) approach. This method use to perform the analysis of product selling time in e-commerce. The samples were collected by survey on website. The results areMARS model with martingale residuals has good performance than residual deviance. MARS modeling with martingale residuals have GCV minimum 0.502 with a combination of BF = 10, MI = 1, and MO = 2 with information number of products sold (X6) that contribute. Variables significant effect on α = 5% were BF_2 = (X_6-135)+, BF_3 = (X_6-170)+, and BF_5=(X_6-196)+.