Telco customer churn IBM dataset
Description
This dataset is a globally recognized, real-world customer churn benchmark originally provided by IBM Sample Data Sets to support predictive modeling, retention analytics, and customer relationship management (CRM) research. It captures an enterprise-level cross-sectional snapshot of 7,043 telecommunications subscribers in California. Each record represents an individual customer, detailing their demographic profile, account configurations, financial commitments, and active service tier subscriptions. Within empirical research pipelines, this dataset serves as an ideal baseline for evaluating cross-domain model generalizability, testing out-of-domain transportability, and benchmarking static feature-space classifiers against advanced sequence-based models. Data Schema & Feature Definitions The raw data matrix consists of 21 columns (features) categorized into four core operational dimensions: Target Variable: Churn (Binary/Nominal: Yes/No), indicating whether the customer discontinued their service subscription within the last month. Account & Financial Metrics: customerID (Unique identifier), tenure (Total months stayed with the company), Contract (Type: Month-to-month, One year, Two year), PaperlessBilling (Yes/No), PaymentMethod (Electronic check, Mailed check, Bank transfer, Credit card), MonthlyCharges (Continuous), and TotalCharges (Cumulative financial expenditure). Core & Ancillary Service Subscriptions: Tracks structural line and infrastructure add-ons: PhoneService (Yes/No), MultipleLines (Yes/No/No phone service), InternetService (DSL, Fiber optic, No), and six specific tech add-ons: OnlineSecurity, OnlineBackup, DeviceProtection, TechSupport, StreamingTV, and StreamingMovies (all structured as Yes/No/No internet service). Demographic Attributes: Basic subscriber profiles including gender (Male/Female), SeniorCitizen (Binary: 1 = Yes, 0 = No), Partner (Yes/No), and Dependents (Yes/No). Research Context & Utility This dataset is highly valuable for validation frameworks evaluating distribution shifts and model transportability. While it represents a cross-sectional structure where temporal behaviors are aggregated into static indicators (such as overall tenure and total financial charges), it provides pristine joint statistical dependencies mapping service configurations directly to attrition intent. Researchers leveraging complex temporal sequential models can aggregate their sequential outputs to a static, feature-collapsed vector space to evaluate whether the latently discovered behaviors retain predictive structural equivalence when deployed against this standard IBM enterprise baseline.
Files
Institutions
- Beijing University of Posts and TelecommunicationsBeijing, Beijing