AI-Driven Predictive Maintenance and Quality Assurance of Enterprise Banking Software Systems
Muddassar Sattar Alvi
U.E.T Lahore
Keywords: predictive maintenance, software quality assurance, AIOps, anomaly detection, defect prediction, banking software, MLOps, operational resilience
Abstract
Enterprise banking platforms (core banking, payments, and channel systems) must sustain very high availability while shipping frequent changes under strict supervisory expectations for operational resilience. Reactive operations and manual quality assurance (QA) struggle to meet both demands. This paper proposes a four-layer, governance-aware framework that couples AI-driven predictive maintenance (PdM)—anomaly detection and failure-horizon prediction on telemetry—with AI-driven QA—change-risk prediction and risk-based test allocation. We synthesise the 2020–2026 literature on industrial PdM, AIOps, log analytics, software defect prediction, LLM-assisted testing, and ML operations, and identify gaps specific to regulated banking environments. The framework is then evaluated through a controlled, fully synthetic simulation study: 40 service instances over 100 days of 30-minute telemetry containing 612 injected incidents, and 12,000 time-ordered code changes. On the held-out period, a random forest raised F1 from 0.562 (static threshold rules) to 0.760, extended mean warning lead time from 67 to 103 minutes, and reduced false-alert episodes by 38%; an unsupervised Isolation Forest detected incidents but generated 2.26 false-alert episodes per instance-week. Models caught every gradual-degradation incident but almost no sudden failures (3%), bounding achievable event recall. Rolling retraining offered only modest benefit after a simulated platform migration. In QA, learned change-risk models found 2.3 times more defects than a size-only baseline within 20% of review effort. Because the data are synthetic, results demonstrate relative behaviour and methodological trade-offs, not production performance; we provide a validation roadmap for real bank environments.
