Learning from Change: Predictive Models for Incident Prevention in a Regulated IT Environment

2022-12-27 · Source: cs.AI updates on arXiv.org · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Data Science & Analytics, Software Development & Engineering · Depth: Expert, extended

Summary

A study conducted at a large international bank, ING, presents a predictive incident risk scoring approach designed to prevent IT incidents caused by changes in highly regulated financial environments. The research compares the bank's existing rule-based risk assessment with three machine learning models: HGBC, LightGBM, and XGBoost, using a one-year dataset of 175,000 closed change tickets and linked priority 1 and 2 incidents. LightGBM emerged as the best-performing model, especially when enriched with aggregated team metrics like change success rates and incident counts. The approach emphasizes auditability and explainability, utilizing SHAP values to provide feature-level insights and ensure transparent, traceable decisions, which is critical for regulatory compliance under frameworks like DORA and the EU AI Act. This data-driven method significantly outperforms the baseline rule-based system in identifying high-risk changes, enabling proactive risk mitigation and enhancing IT operational reliability.

Key takeaway

For CTOs and VPs of Engineering managing IT change in regulated sectors, adopting data-driven ML models like LightGBM for incident prediction offers superior risk assessment compared to traditional rule-based systems. You should integrate explainable AI (XAI) techniques, such as SHAP, into your change management workflows to ensure auditability and build trust. This enables proactive identification of high-risk changes, allowing your teams to apply targeted scrutiny and preventive actions, thereby enhancing operational resilience and compliance with regulations like DORA and the EU AI Act.

Key insights

Interpretable ML models can effectively predict IT incident risk from changes in regulated environments, outperforming rule-based systems.

Principles

• Explainability is crucial for regulatory compliance and user trust.
• Aggregated team metrics enhance predictive model accuracy.
• High recall is prioritized over precision in incident prevention.

Method

Train boosted tree classifiers (LightGBM) on historical change and incident data, enriched with aggregated team metrics. Use SHAP for feature-level interpretability to provide incident prediction scores for planned IT changes.

In practice

• Use SHAP to explain model predictions to engineers.
• Incorporate team performance metrics into risk models.
• Prioritize recall to identify most incident-inducing changes.

Topics

• Predictive Incident Prevention
• IT Change Management
• Explainable AI
• LightGBM Model
• Financial Sector Regulation

Best for: CTO, VP of Engineering/Data, Director of AI/ML, Machine Learning Engineer, MLOps Engineer, Research Scientist

Related on AIssential

• Learning from Change: Predictive Models for Incident Prevention in a Regulated IT Environment — Interpretable machine learning models can enhance IT change management and incident prevention in regulated environments.
• Managing Third-Party Risk at Scale Without Drowning in Surveys - with Carey Smith — Continuous, AI-enabled monitoring is essential for managing multi-tier supply chain risks at scale.
• Presentation: DevOps Modernization: AI Agents, Intelligent Observability and Automation — AI agents are transforming DevOps by enabling predictive operations, intelligent rollouts, and machine-speed incident remediation.
• Using observability data to prevent incidents — Proactive reliability intelligence, enabled by rapid data access, prevents incidents by identifying risk trends before they manifest.
• AI can meaningfully improve healthcare planning—but only when it is transparent, clinically grounded, and operationally embedded. — Explainable AI models can accurately predict hospital stay and costs, with LOS being the primary cost driver.
• Token-Efficient Change Detection in LLM APIs — B3IT enables cost-effective, black-box LLM change detection using specific "Border Inputs" that yield multiple top output tokens.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by cs.AI updates on arXiv.org.