Harnessing Machine Learning to Mitigate Cyber Risks in Construction

In the ever-evolving construction industry, machine learning (ML) is now being harnessed to address the critical issue of cyber risks. A groundbreaking approach has been developed to predict and mitigate the cyber threats that construction projects frequently encounter.

This innovative framework is built on three core components: a Monte Carlo-simulated dataset for predicting cyber risks, an ML-driven analysis to pinpoint key risk factors, and a greedy optimization algorithm designed to prioritize and effectively manage high-risk elements.

Understanding the Context

While digital technologies have greatly improved efficiency and productivity in construction, they have also introduced new cybersecurity vulnerabilities. These vulnerabilities can result in project delays, financial losses, and damage to reputations. Compared to other sectors, the construction industry has been slower to adopt comprehensive cybersecurity measures, leading to a significant rise in cyber incidents over the last ten years.

Construction projects are primarily threatened by five types of cyber risks: ransomware attacks on critical assets, phishing schemes, insider threats, data breaches, and supply chain attacks. To counter these threats, project managers need predictive tools that can forecast risks throughout a project's lifecycle. Such tools enable proactive risk mitigation, ensuring projects proceed smoothly.

Methodological Approach

The development of this ML-centric approach involved a structured, multi-step process. Initially, feature sources for the ML models were identified based on construction-specific risk factors. Datasets were then generated using Monte Carlo simulations and an ensemble labeling method, ensuring robust datasets for model training.

The model development followed a two-phase strategy. The first phase identified the best-performing ML model for each risk, while the second phase optimized weight combinations for different labeling methods. An ML-based feature analysis was then conducted to identify significant risk factors, followed by the development of a greedy optimization algorithm to create effective risk reduction strategies.

The outcome was a dynamic cyber risk assessment tool comprising three key modules: trained ML models for risk prediction, a risk factor analysis module, and a risk reduction strategy module. Together, these modules offer a comprehensive framework for analyzing and mitigating construction risks.

Findings and Implications

The study found that insider attack risk consistently achieved high determination coefficient (R²) values across all ML models, indicating a linear relationship. Simpler models were deemed suitable for predicting these risks. In contrast, ransomware, phishing, and data breach risks exhibited non-linear relationships, requiring more complex models for accurate prediction.

These findings highlight the complexity of cyber risks in construction and the necessity for project managers to grasp these intricacies. Each type of risk presents unique non-linear relationships, suggesting that tailored strategies are essential for effectively addressing each cyber threat.

The ML models proved their effectiveness by accurately predicting cyber risks in both expert-labeled projects and a real construction project. This capability enables project managers to implement immediate risk reduction strategies, guided by the model’s optimization algorithm to maximize resource allocation efficiency.

Concluding Thoughts

The study successfully developed an ML-based approach to assess common cyber risks in construction, including ransomware, insider threats, data breaches, phishing, and supply chain attacks. Despite challenges such as the lack of existing datasets, the researchers created a simulated dataset using defined probability distributions, validated by experts.

Future efforts will focus on refining these probability distributions through sensitivity analyses and expanding expert reviews to enhance simulation accuracy. Collaborations with local companies to gather real-world data will further validate the models, bridging the gap between theoretical modeling and practical application.