Answers are grounded in this paper and cite their sources.
The volume of digital payment transactions in Indonesia is very high. Large-scale disruptions and degradation of QRIS transaction services have frequently caused financial losses. Therefore, QRIS operational services require robust, adaptive anomaly detection. This study evaluates five unsupervised learning algorithms. The algorithms employed are Unsupervised k-Nearest Neighbors (k-NN), Isolation Forest (iForest), Local Outlier Factor (LOF), One-Class Support Vector Machine (OCSVM), and Autoencoder. A combination of the total_suspect_log1p and divergence features is explored and derived from the characteristics of suspect transactions and failed transactions. Conventional thresholds are often unable to capture anomalies. Therefore, Median Absolute Deviation (MAD) is employed to construct a ground-truth proxy as a reference for evaluation. Testing was conducted in mid-June and mid-July 2026 to provide a six-month retrospective and prospective view of transaction patterns. They also cover the Eid al-Fitr period in 2026, when transaction surges occurred. The results show that the Unsupervised k-Nearest Neighbors algorithm outperforms the other algorithms, achieving an F2-score of 92.53 and a recall of 93.16. These findings show that selecting anomaly detection algorithms and features should account for the characteristics of QRIS transaction data. This study contributes to the development of a monitoring system for QRIS transaction service degradation to minimize system downtime.
No takes yet. Share an insight, caveat, or question.
Kemando et al. (2026) studied this question.
Synapse has enriched 8 closely related papers on similar clinical questions. Consider them for comparative context: