ENHANCING OCCUPATIONAL SAFETY IN CONSTRUCTION THROUGH INTEGRATED RISK MODELLING, REAL-TIME EVENT SOURCING, AND AI-ASSISTED DECISION SUPPORT: A MULTIDISCIPLINARY FRAMEWORK

Abstract

Background: Construction remains one of the highest-risk economic sectors worldwide, characterized by complex spatio-temporal hazard exposure, heterogeneous workforce profiles, and fragmented information flows (Sacks et al., 2009; International Labour Organization, 2023). Traditional safety analysis techniques—while foundational—struggle to accommodate high-velocity data streams and cross-scale interactions that produce many modern accidents (Macedo & Silva, 2005; Zhang et al., 2019). Recent advances in machine learning, event sourcing architectures, and decision support systems offer potential to bridge these gaps, yet there is limited integrative scholarship that ties occupational statistics, engineering hazard analyses, and contemporary computational approaches into an actionable safety framework (Eurostat, 2021; Kang & Ryu, 2019; Kesarpu & Dasari, 2025).

Objective: This paper develops a comprehensive, publication-ready, original research article that synthesizes statistical evidence on construction accidents, theoretical insights into spatial–temporal hazard exposure, and applied methods from machine learning and real-time event sourcing to propose an integrated framework for enhancing safety and health outcomes on construction sites. The framework is designed to be implementable within prevailing industry information systems and to be sensitive to the workforce, regulatory, and technological constraints documented across regions (International Labour Organization, 2023; OSHA, 2023; Eurostat, 2024).

Methods: We employ a multi-method conceptual and applied approach grounded in four pillars: (1) rigorous descriptive synthesis of occupational accident statistics and cross-national comparisons (Eurostat, 2021; Choi et al., 2019); (2) theoretical expansion of spatial and temporal exposure models derived from published hazard mapping and site dynamics (Sacks et al., 2009; Cabello et al., 2021); (3) methodological mapping of machine learning models appropriate for predicting accident types and severity, including random forests and ensemble approaches (Kang & Ryu, 2019; Pillai, 2023); and (4) architectural design for real-time event sourcing (Kafka) to support low-latency risk analysis and AI-assisted decision support (Kesarpu & Dasari, 2025; Dunka, 2022). Each pillar is elaborated with operational detail to allow replication and adaptation.

Results: The integrated framework specifies data schemas, model selection rationales, risk aggregation strategies, and human-in-the-loop decision pathways. Descriptive analysis of the literature shows consistent injury patterns by task and profession, differential fatality profiles across countries, and strong associations between exposure timing and accident clustering (Macedo & Silva, 2005; Choi et al., 2019; Cabello et al., 2021). Machine learning suitability is demonstrated through argumentation: classification models (e.g., random forest) for accident type prediction, survival/ hazard-like approaches for time-to-event risk estimation, and unsupervised methods for anomaly detection in sensor streams (Kang & Ryu, 2019; Pillai, 2023). Event-sourced pipelines map sensor, schedule, and worker-reported events into immutable logs feeding real-time feature extraction and model inference (Kesarpu & Dasari, 2025).

Conclusions: Combining descriptive occupational statistics with spatial–temporal hazard theory and modern computational architecture yields a viable path to materially improving safety outcomes. Challenges remain—data governance, workforce acceptance, and regulatory harmonization—but the framework provides clear technical and organizational steps for piloting and scaling interventions across diverse regulatory contexts (International Labour Organization, 2023; OSHA, 2023). This research contributes an integrative theoretical and operational blueprint linking epidemiological evidence and computational systems to practical on-site risk reduction.