Comparative Analysis of Probabilistic Risk Models for Sustainable Construction Risk Management in Developing Economies

This repository contains anonymized survey data, longitudinal project observations, probabilistic model outputs, statistical validation results, and hybrid AI–Markov framework performance metrics supporting the comparative evaluation of probabilistic risk assessment models for sustainable construction risk management in infrastructure projects within Nigeria’s Niger Delta. The dataset comprises quantitative survey responses from 442 construction professionals, 775 longitudinal risk-state transition observations from eight infrastructure projects between 2020 and 2024, expert elicitation data from 23 specialists, and comparative performance metrics for Markov Chains, Bayesian Networks, Monte Carlo Simulation, and Bow Tie Analysis. The study evaluated the relative performance of these probabilistic models across predictive accuracy, computational efficiency, dynamic adaptability, stakeholder interpretability, and technology integration potential. Consistent with the study hypothesis, the research examined whether statistically significant differences existed among the selected probabilistic models rather than assuming that any single model would consistently outperform the others. Survey data were collected using a validated 30-item structured questionnaire (Cronbach’s α = 0.87), while transition probabilities were estimated from project monitoring and expert consensus through a modified Delphi process. Analytical procedures included Markov Chain modelling, Bayesian inference, Monte Carlo simulation, Bow Tie Analysis, one-way ANOVA, regression analysis, five-fold cross-validation, and multi-criteria performance assessment. The dataset demonstrates that each probabilistic modelling approach exhibits distinct strengths within different decision contexts. Markov Chains achieved the highest temporal prediction accuracy (94.2%) for forecasting project risk-state transitions, Bayesian Networks demonstrated superior adaptability to changing risk conditions and emerging digital technologies, Monte Carlo Simulation provided probabilistic estimates of cost and schedule uncertainty, and Bow Tie Analysis exhibited the highest stakeholder interpretability despite comparatively lower predictive performance. The accompanying hybrid AI–Markov framework further improved predictive performance by reducing forecast error by about 40% and explaining 77.9% of the observed variation in project sustainability performance. The dataset supports reproducible research, comparative evaluation of risk models, and the development of advanced risk management methodologies for sustainable infrastructure delivery. It may be reused for benchmarking predictive models, validating machine learning algorithms, conducting sensitivity and uncertainty analyses, developing hybrid decision-support frameworks, and supporting evidence-based risk management in construction and infrastructure projects operating under complex environmental, socioeconomic, and institutional conditions.