Analyzing Customer Churn Prediction in E-Commerce Using Survival Analysis Techniques

Thesis Abstract

In the rapidly evolving landscape of e-commerce, customer retention remains a critical determinant of organizational success, yet the propensity for customers to cease engaging with online platforms—termed customer churn—poses significant challenges in sustaining competitive advantage. This study seeks to develop a robust analytical framework for predicting customer churn by employing survival analysis techniques, thus providing strategic insights to enhance customer retention strategies within the e-commerce industry. The primary aim is to investigate the temporal dynamics of customer attrition and identify key factors influencing the duration until churn, with a focus on leveraging survival analysis models to improve prediction accuracy. To achieve this aim, the study formulates specific objectives (1) to examine the patterns of customer engagement and churn in an e-commerce platform over a 12-month period; (2) to identify significant demographic, behavioral, and transactional predictors of customer churn; (3) to compare the performance of various survival analysis models—such as the Kaplan-Meier estimator, Cox proportional hazards model, and accelerated failure time (AFT) models—in predicting churn timing; and (4) to develop an optimized predictive model for customer lifetime, integrating these variables for practical application. The research adopts a quantitative, longitudinal case study design. The population comprises active registered customers of a major e-commerce retailer operating within North America, totaling approximately 500,000 users. Employing stratified random sampling, a sample size of 2,000 customers is determined based on Cochran’s formula, with stratification to ensure representative coverage across segments such as age, purchase frequency, and geographic location. Data collection utilizes a combination of secondary transactional data from the retailer’s customer relationship management (CRM) system and primary data gathered through structured online surveys designed to capture customer engagement behaviors, satisfaction levels, and demographic information. Data quality is ensured through validity and reliability assessments, including Cronbach’s alpha for internal consistency of survey instruments. Data analysis involves descriptive statistics to profile customer behavior, followed by non-parametric Kaplan-Meier survival estimates to visualize customer retention rates over time. Inferential analysis employs Cox proportional hazards regression to identify and quantify the effect of predictor variables, with goodness-of-fit assessed via log-likelihood ratio tests. Accelerated failure time models are also employed to compare model performances, utilizing criteria such as Akaike Information Criterion (AIC) and Bayesian Information Criterion (BIC). The study incorporates proportional hazards assumptions testing and model validation through residual analysis and cross-validation techniques to ensure robustness. Expected findings suggest that factors such as purchase frequency, customer satisfaction scores, average transaction value, and demographic variables significantly influence customer churn timing, with survival models providing nuanced insights into retention durations. The comparative analysis is anticipated to demonstrate that the Cox proportional hazards model offers superior predictive power within this context, though AFT models may reveal additional insights into the factor effects on churn time. This research contributes to the body of knowledge by applying and comparing advanced survival analysis techniques within the e-commerce domain, filling existing gaps regarding dynamic prediction of customer attrition. It offers a practical predictive framework for ecommerce managers aiming to identify at-risk customers proactively and implement targeted retention interventions. The study concludes with recommendations for integrating survival analysis models into real-time customer relationship management systems and suggests avenues for future research, including the incorporation of machine learning algorithms to enhance predictive accuracy and adaptability in changing market conditions.