WSDM - KKBox's Churn Prediction Challenge_transactions_v2.csv.7z
Description
This file is a vital component of the benchmark dataset donated by KKBOX (Asia’s leading music streaming service) for the WSDM (ACM International Conference on Web Search and Data Mining) Churn Prediction Challenge. The dataset contains real-world, granular, longitudinal transactional logs of multi-period user interactions up to March 31, 2017. Unlike traditional cross-sectional customer datasets, this file records dynamic behavioral trajectories—including sequential updates to subscription lengths, plan downgrades/upgrades, pricing changes, and explicit cancellation actions. Data Schema & Feature Definitions The schema contains 9 structural variables tracking temporal customer behavior and transactional states: 1. 'msno' (Object/String): The unique, encrypted, anonymized user identifier used to align transactional sequences to the main user profile and target churn matrices. 2. `payment_method_id` (Integer/Factor): An encoded indicator representing the specific platform or billing method utilized for the transaction (e.g., credit card, mobile billing, convenience store pay). 3. `payment_plan_days` (Integer): The formal duration of the chosen membership subscription plan, measured explicitly in days (e.g., 7 days, 30 days, 180 days). 4. `plan_list_price` (Integer): The standard retail or listed price of the subscription plan, denominated in New Taiwan Dollars (NTD). 5. `actual_amount_paid` (Integer): The actual net financial amount paid by the customer in New Taiwan Dollars (NTD). Discrepancies between this and the `plan_list_price` capture promotional discount depths. 6. `is_auto_renew` (Binary/Boolean): A flag (`1` = Yes, `0` = No) indicating whether the subscription plan is configured to automatically renew at the next expiration boundary. 7. `transaction_date` (Integer/Date): The precise timestamp of the transaction execution, captured in `%Y%m%d` format. 8. **`membership_expire_date`(Integer/Date): The calculated expiration timestamp for the user's valid membership tier resulting from the transaction, captured in `%Y%m%d` format. 9. `is_cancel` (Binary/Boolean): A flag (`1` = Yes, `0` = No) capturing whether the user explicitly triggered a manual cancellation of their active membership during this specific transactional event. (Note: A value of 1 does not necessarily mean immediate churn but rather indicates a structural shift in user intent). Research Context & Utility This file is uniquely structured for time-series, survival analysis, and deep sequential neural networks (e.g., Recurrent Neural Networks, LSTMs, and Transformer Encoder architectures). Because an individual user (`msno`) can generate dozens of rows over time as they modify their subscription states, researchers can construct sequential time-stamped paths. This allows for advanced modeling of the velocity of behavior—such as calculating inter-transaction gaps, sequence-level drop-offs in auto-renewal status, and temporal proximity of explicit cancellation actions.
Files
Institutions
- Beijing University of Posts and TelecommunicationsBeijing, Beijing