Indonesian Mobile-Commerce Customer Behavior & Demographics

The paper uses a proprietary, multi-source dataset (no formal public name) compiled from structured surveys and anonymized behavioral logs pulled via m-commerce platform APIs for major Indonesian platforms—Shopee, Tokopedia, Bukalapak, and BliBli. The final sample includes 5,127 valid user records collected between March and August 2024, with sample size set by statistical power analysis to ensure adequate representation across demographics and usage patterns. The cohort skews young (69% aged 18–35), is near gender-balanced (48% male; 52% female), and is geographically concentrated in Java (60%), with Sumatra and other islands each at 20%. The prediction target is a binary churn label modeled with a single-neuron sigmoid output that produces churn probabilities. The input feature set comprises 47 engineered variables covering transactional behavior, engagement metrics, demographic attributes, temporal patterns, and customer-service interactions; preprocessing applies median/mode imputation for missing values, IQR-based outlier screening, and z-score standardization for numeric features.