Machine Learning–Based AML/KYC Transaction Monitoring for Suspicious Activity Detection and Compliance Risk Reduction in Digital Banking

Abstract

This study addresses the persistent problem in digital banking that traditional rule-based AML/KYC transaction monitoring generates high alert noise and inconsistent investigative outcomes, increasing compliance cost and residual regulatory exposure. The purpose was to quantify how machine learning enabled transaction monitoring capabilities contribute to suspicious activity detection and perceived compliance risk reduction in a cloud-enabled enterprise digital banking case context. Using a quantitative cross-sectional, case-based design, data were collected via a structured 5-point Likert survey from professionals embedded in AML operations, compliance oversight, KYC/CDD, risk, and AML technology functions (N = 168 valid responses; mean AML experience = 6.2 years, SD = 2.9; role distribution led by transaction monitoring analysts at 42.3%). Key variables included ML Detection Effectiveness (MDE), Data Quality and Feature Readiness (DQF), , Analyst Trust and Adoption (ATA), and three domain indices: Alert Quality and Workload Impact (AQWII), Explainability Audit Readiness and Model Governance Evidence (EARMGE), and Typology Coverage and Adaptability Results (TCAR), with Compliance Risk Reduction (CRR) as the dependent variable. The analysis plan applied descriptive statistics, reliability testing (Cronbach’s alpha), Pearson correlations, and multiple regression to estimate unique predictors of CRR. Results showed strong overall agreement that ML monitoring improved outcomes (CRR M = 4.14, SD = 0.52; MDE M = 4.18, SD = 0.54; AQWII M = 4.12, SD = 0.55), and all constructs met good to excellent internal consistency (α = 0.83–0.91). CRR correlated most strongly with MDE (r = 0.72, p < .001) and AQWII (r = 0.69, p < .001), indicating that detection strength and workload-relevant alert quality moved together with compliance benefit. The regression model was significant and explanatory (R² = 0.68; F(8,159) = 42.11, p < .001), with headline effects from MDE (β = 0.31, p < .001) and AQWII (β = 0.23, p < .001), alongside meaningful contributions from ATA (β = 0.18, p = .002), DQF (β = 0.15, p = .007), SIA (β = 0.14, p = .015), MET (β = 0.12, p = .029), EARMGE (β = 0.13, p = .018), and TCAR (β = 0.11, p = .042).