.png){kind=icon}
.png){kind=icon}
Return
Machine learning and churn: predicting season ticket holder behaviour
.png){kind=icon}
OA
AI
C
D
A
J
K
DOI：10.1080/14413523.2026.2647540.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Season ticket holder (STH) churn presents a persistent challenge for professional sport organisations. Although machine learning offers considerable potential, prior studies typically rely on cross-sectional datasets or aggregate-level features, limiting their predictive utility. Drawing on customer lifetime value (CLV) theory, this study aims to: develop and evaluate churn-prediction models integrating longitudinal individual-level data, compare the performance and feature importance of multiple machine learning algorithms, and examine how disruptive events, such as the COVID-19 pandemic, influence model performance. Individual-level data were sourced from a single professional sport team in Australia across six seasons (2018–2023), encompassing 118,469 STH decisions. Six machine learning algorithms were compared to assess differences in predictive performance and feature importance across varying temporal windows. Findings indicate that models trained on shorter, two-season longitudinal datasets organised around the disruption, substantially outperformed the full six-season model. The strongest results were achieved using boosted tree algorithms, with CatBoost producing the highest accuracy (0.821 ± 0.002) and AUC-ROC (0.901 ± 0.002) in the post-COVID period. Feature importance analyses revealed a stable set of core predictors of churn (i.e. tenure, driving distance, and late-season attendance) with boosted models showing high internal consistency. The findings have important theoretical and practical implications. Theoretically, the findings are consistent with customer lifetime value frameworks but show that the behavioural signals underpinning retention shift following disruption. Practically, the results demonstrate that sport organisations can build effective churn models using relatively short longitudinal datasets to support targeted, data-driven retention strategies.
Keywords:
Retention
professional teams
prediction
algorithm
disruption
Journal
IF:
3.8
Papers:
1.8K
Citations:
3.2K
Organization
Cited Papers