With the booming of the Internet era, ecommerce has become a major sector and changed the traditional retail industry dramatically with low sales cost and intense marketing events. Massive promotions such as “Single’s Day” created by Chinese ecommerce giant Alibaba have become a widely accepted and practiced competition point.
One problem aroused from this situation is the effectiveness and efficiency of these marketing events, which constitute a large proportion of online retailer’s costs. A successful marketing event would expect to convert new customers into repeated buyer eventually in his or her life cycle instead of one-time discount hunter. Therefore prediction on future consumer behavior becomes essential in order to evaluate and improve these events. Fortunately, one of the new features of ecommerce is the rich data pool on valuable information such as demographic profile, interactive behavior, transaction history, etc., which can be used to learn and predict whether a customer will be converted into repeated buyer for certain merchant through certain event. With a well-trained model, online retailers will be able to target more precisely on high conversion probability customers therefore improve the marketing efficiency and business profitability in general.
Thus, the aim of this project is to generate and evaluate some of the possible models based on the given data and to create a repeatable learning process for future use of any other stakeholders. Probabilistic generative process will be applied on customer features include demography, interaction, and purchase history to give a symbolic probability of conversion and then selected decision rule will classify this given customer into one of the two categories: conversion and non-conversion, which are represented by the predicting target label 1 and 0.