Customer Targeting in E-Commerce: A Feature Selection and Machine Learning Approach

Abstract

For e-commerce and web-based businesses, being able to identify which of their online visitors are likely to convert to customers is a top priority that has significant implications for ad-targeting and profitability. In this paper, a customer targeting model is developed to help solve that problem. The model takes in, as input data, a database that contains records of individuals and features that describe the demographic attributes and browsing behavior of that individual. The first phase of the model reduces the dimensionality of the database through feature selection to arrive at an optimal feature subset. The second phase of the model uses a machine learning based classifier to classify visitors into customers and non-customers based on the optimal feature subset. 26 unique model implementations that pair different feature selection and machine learning algorithms for the 2 phases were tested on a real input database acquired from an e-commerce start-up. Most of the model implementations significantly outperform naive classification but only the 2 best implementations could match human learning and classification. Additionally, it was found that hybrid models that combine machine and human learning were able to perform as well as the best standalone human learning and the best standalone machine learning models.