Simulation-Based Framework for Segmenting QRIS Consumer Payment Patterns Using Transaction Analytics and K-Means Clustering

The Quick Response Code Indonesian Standard (QRIS) has accelerated the adoption of cashless retail payments in Indonesia, yet the consumers behind these transactions exhibit heterogeneous payment patterns that remain difficult to study because individual-level transaction records are restricted by confidentiality, ownership, and privacy constraints. This study develops and evaluates a simulation-based framework for segmenting QRIS consumer payment patterns using transaction analytics and K-Means clustering. A synthetic dataset of 20,686 transaction records for 1,000 synthetic users across 50 merchants in seven merchant categories was generated over a 180-day period from four explicitly planted latent behavioral profiles, with controlled noise comprising failed transactions, duplicates, missing categories, and extreme values. After preprocessing, 20,277 valid records were aggregated into eight user-level behavioral features covering frequency, monetary value, merchant diversity, temporal concentration, and regularity. Combined evidence from the Elbow curve, silhouette coefficient, Davies–Bouldin Index, and Calinski–Harabasz Index supported K = 4. The resulting segments Frequent Low-Value, Occasional High-Value, Diverse Active, and Emerging users recovered the planted profiles with an Adjusted Rand Index of 0.947, Normalized Mutual Information of 0.920, and an overall matching proportion of 97.6%, and remained stable across 30 repeated runs (mean ARI = 1.000) and six sensitivity scenarios (ARI 0.941–0.988). The framework offers a transparent, reproducible analytical workflow that can later be applied to anonymized real transaction data. Because the data are synthetic, the segments demonstrate methodological performance under stated assumptions and must not be generalized to actual QRIS users in Indonesia.