Machine Learning Models for Bank Reconciliation Anomaly Detection in High-Volume Financial Transactions

Abstract

This study examined the role of machine learning models in improving bank reconciliation anomaly detection in high-volume financial transaction environments, where manual review, spreadsheet-based matching, and fixed rule-based systems often fail to identify duplicate entries, unmatched records, amount mismatches, missing references, timing delays, failed payments, reversals, and suspicious transaction patterns quickly enough. The purpose of the study was to determine whether machine learning model capability, transaction data quality, automated anomaly detection capability, reconciliation system automation, and organizational readiness significantly influence bank reconciliation anomaly detection performance. A quantitative, cross-sectional, case-based research design was adopted, using cloud-supported and enterprise financial operation cases involving banks, fintech firms, payment processors, accounting units, audit departments, risk teams, compliance units, and financial control professionals. Data were collected through a structured five-point Likert-scale questionnaire from 204 valid respondents out of 250 distributed questionnaires, producing an 81.6% valid response rate. The key variables included machine learning model capability, transaction data quality, automated anomaly detection capability, reconciliation efficiency, operational risk reduction, financial control reliability, organizational readiness, environmental pressure, and bank reconciliation anomaly detection performance. The analysis plan involved descriptive statistics, Cronbach’s Alpha reliability testing, Pearson correlation, multiple linear regression, anomaly typology ranking, maturity index assessment, and hypothesis testing using SPSS. The findings showed strong quantitative support for the study model: automated anomaly detection capability recorded the highest mean score of 4.22, followed by bank reconciliation anomaly detection performance at 4.20 and machine learning capability at 4.18. Reliability was strong, with an overall Cronbach’s Alpha of 0.91. Correlation results showed significant positive relationships between anomaly detection performance and automated anomaly detection capability, r = 0.72, machine learning capability, r = 0.68, operational risk reduction, r = 0.66, transaction data quality, r = 0.63, and financial control reliability, r = 0.61. Regression results confirmed that the model explained 62.4% of the variance in anomaly detection performance, R² = 0.624, F = 54.27, p < 0.001, with automated anomaly detection as the strongest predictor, β = 0.31, followed by machine learning capability, β = 0.26, and transaction data quality, β = 0.22. The study implies that banks and enterprise financial systems should integrate machine learning-based anomaly detection with high-quality transaction data, automated workflows, staff training, governance controls, and audit-ready reporting to improve reconciliation efficiency, reduce operational risk, and strengthen financial control reliability.