Mining customer behavior in trial period of a web application usage—case study

Abstract

This paper proposes models for predicting customer conversion from trial account to full paid account of web application. Two models are proposed with focus on content of the application and time. In order to make a customer’s behavior prediction, data is extracted from web application’s usage log in trial period and processed with data mining techniques. For both models, content and time based, the same selected classification algorithms are used: decision trees, Naïve Bayes, k-Nearest Neighbors and One Rule classification. Additionally, a cluster algorithm k-means is used to see if clustering by two clusters (for converted and not-converted users) can be formed and used for classification. Results showed high accuracy of classification algorithms in early stage of trial period which can serve as a basis for an identification of users that are likely to abandon the application and not convert.