Adaptive Machine Learning for SME Financial Distress and Transaction Fraud Detection

Abstract

Small and medium-sized enterprises (SMEs) present a harder machine learning problem than either large corporate borrowers or individual consumers, for reasons the credit-risk and fraud-detection literatures document separately: SMEs are financially heterogeneous across industries and life-cycle stages in ways a single static scorecard struggles to generalize across, their business-banking transaction patterns shift rapidly as the enterprise itself grows, contracts, or changes activity, and a lender or business-banking provider typically monitors both an SME’s solvency and its transaction integrity through separate systems that do not share what either one learns. A substantial peer-reviewed literature shows that financial-ratio and non-financial information can predict SME distress with useful lead time, that machine learning classifiers benchmark well against traditional scorecards for credit risk more broadly, and that a model’s accuracy decays as the population it scores drifts away from the population it was trained on — yet no existing framework treats SME financial-distress prediction and SME transaction-fraud detection as a single problem requiring the same underlying defense against exactly this kind of drift. This paper addresses that gap by developing Sequential Model Adaptation for Risk and Transaction-fraud (SMART): a reference architecture in which a shared, continuously monitored scoring engine produces both an SME financial-distress score and a transaction-fraud score from the same enriched SME data, a dedicated concept-drift detection layer triggers targeted model updates precisely when and where the incoming data or model performance indicates drift, rather than on a fixed retraining schedule, and an explainability layer attaches a governance-ready rationale to every score reaching a relationship manager or risk officer. Because no production deployment of this specific architecture exists to evaluate, the paper follows a design-science approach: the framework is built from and evaluated against peer-reviewed findings in SME financial distress and failure prediction, relationship-based and non-financial information in SME credit risk, concept drift and adaptive learning for evolving financial data, machine learning benchmarking for credit risk classification, transaction and business-banking fraud detection, and explainability and fair-lending considerations in adaptive SME risk and fraud systems, and every quantitative comparison presented — two tables and four  figures spanning capability coverage, a twelve-month accuracy trajectory under population drift, a drift-magnitude-to-detection-lag relationship, and relationship-manager review-hour reallocation by SME segment — is explicitly labeled as a literature-grounded  estimate rather than a measured result. The paper’s contribution is a synthesis of six otherwise separate literatures into a single adaptive architecture that treats SME distress prediction and SME transaction-fraud detection as jointly exposed to the same drift problem, and a candid account of where adaptive machine learning helps an SME-facing lender or business-banking provider.