A Comparison Study of Parametric and Machine Learning Survival Analysis Models to Predict Customer Churn in the Edtech Sector - Master’s Thesis

Aug 5, 2026

Upload by Eden

This thesis explores the application of survival analysis models to predict customer churn in the edtech sector, where subscription-based retention is a core growth driver. It leverages statistical and machine learning methods to improve retention modeling beyond heuristic baselines and to determine which variables shape churn dynamics. Using a survival framework suitable for censored data, the work targets churn likelihood and retention duration with more precise, actionable outputs. Experiments on a dataset of several hundred thousand customers evaluate classical approaches and advanced model families, including time-variant feature handling, and assess performance via concordance and integrated Brier scoring.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages82

Reading Time3 h 27 min

AccessFree to download

1 Introduction

1.1 Motivation and Problem Statement

1.2 Aim of the Work

1.3 Structure of this Thesis

2 Theoretical Background

2.1 Customer Analytics in Subscription Businesses

2.2 Survival Analysis

2.3 Survival Analysis Models

2.4 Evaluation Metrics for Survival Analysis

2.5 Incorporation of Time-Variant Data

3 Machine Learning for Survival Analysis

3.1 Classic ML Methods

A Comparison Study of Parametric and Machine Learning Survival Analysis Models to Predict Customer Churn in the Edtech Sector MASTER’S THESIS submitted in partial fulﬁllment of the requirements for the degree of Master of Science in Data Science by Benjamin Lee, BA/BE(Hons) Registration Number 12112693 to the Faculty of Informatics at the TU Wien Advisor: Univ.-Prof. Dipl.-Ing. Dr.techn. Peter Filzmoser Vienna, 27th January, 2025 Benjamin Lee Peter Filzmoser Technische Universität Wien A-1040 Wien Karlsplatz 13 Tel. +43-1-58801-0 [www.tuwien.at](www.tuwien.at) Declaration of Authorship Benjamin Lee, BA/BE(Hons) I hereby declare that I have written this Masters Thesis independently, that I have completely speciﬁed the utilized sources and resources and that I have deﬁnitely marked all parts of the work-including tables, maps and ﬁgures-which belong to other works or to the internet, literally or extracted, by referencing the source as borrowed. Vienna, 27th January, 2025 Benjamin Lee iii Acknowledgements It is a cliché, and technically not fully accurate to talk here about nani gigantum humerisinsidentes, but this thesis would not be possible at all were it not for all the people who helped and supported me through this academic journey. First and foremost, my supervisor Dr Peter Filzmoser whose guidance and expertise were invaluable in writing this thesis. Not only did he support me in ﬁnessing the work, but also provided counsel and direction when my literature searches came up blank. I would also want to take the time to acknowledge and thank the examiners who will be reading this thesis and participating in my defensio. Your time and patience is appreciated. To GoStudent and the Finance team, especially Alexander Schaﬀgotsch and Paul Krall, thank you for your support in this endeavour and generosity in providing the data that made this work possible. Michael Weingartner at the Department of Computational Statistics has been an absolute star at setting me up with the hardware system I used for this thesis and I appreciate his patience at dealing with my many questions, particularly so close to Christmas. In my literature review, I would particularly like to thank Luca Badolato, Nicholas Irons, and Weichi Yao for their kindness in answering questions seeking clariﬁcation and expansion on some aspects of the papers they had written. Of course, many other people not directly part of this thesis have also made this academic journey far more bearable and even joyful at times. To my friends and everyone with whom I talked about this thesis, thank you from the bottom of my heart. Special acknowledgements are necessary to the Fachschaft crew, who were my ﬁrst friends in Vienna and still great friends today. In particular, I would like to give a shoutout to: Alicja, Bettina, Christian, Elisabeth, Fränzi, Gunnar, Ivan, Łukasz, Valentin B, and Valentin M. Most important of all, I’d like to thank my family: my Mum, my sister Amanda, and my darling wife Marieke. Your love and belief in me have helped me not only in this journey, but through my life. I love you. Vienna, 27th January 2025 v Abstract This thesis explores the application of survival analysis models to predict customer churn in the edtech sector, an area of growing importance for subscription-based businesses. By leveraging statistical and machine learning techniques, the study aims to improve retention models over existing heuristic methods and identify key variables inﬂuencing churn behaviour. The research focuses on using survival analysis, a statistical framework adept at handling censored data, to predict customer churn and retention duration, providing more precise and actionable insights. Drawing from a dataset comprising several hundred thousand customer records with both time-variant and time-invariant features, this study evaluates classical survival models, including Kaplan-Meier and Cox P

FAQs

What problem does the thesis address in the edtech sector?

It addresses predicting customer churn and retention duration for subscription-based edtech businesses using survival analysis methods.

Which model types are compared in the study?

The study evaluates classical survival models such as Kaplan-Meier and Cox Proportional Hazards, alongside machine learning approaches including Random Survival Forests and Gradient Boosting Machines.

How does the thesis incorporate time-variant information?

It incorporates time-variant features as an additional modeling component, improving model sophistication and predictive capability, while emphasizing the need for correct interpretation.

Aug 5, 2026

Upload by Eden