A research team investigating workplace safety management in high-risk energy infrastructure has introduced and verified a multi-criteria evaluation model termed the Contractor Health, Safety, Environment, and Quality Performance Index, designed to systematically quantify contractor risk in industrial operations. Published in the journal Discover Analytics, the investigation addresses persistent vulnerabilities in the oil and gas sector, where host organizations routinely outsource complex technical activities to independent contractors exhibiting highly disparate safety cultures, training levels, and compliance protocols. Conventional oversight mechanisms in this industry have historically relied on static compliance checklists, minimum qualification thresholds, and lagging metrics such as lost-time injury frequency rates and total recordable incident rates. While these traditional instruments verify baseline regulatory adherence, they offer minimal predictive insight into emerging operational risks and consistently fail to capture the complex, bidirectional interdependencies among organizational, procedural, and behavioural safety dimensions. To resolve this systemic shortcoming, the researchers developed an integrated socio-technical safety governance framework combining structural multi-criteria prioritization with empirical machine learning validation.
The conceptual underpinning of the performance index merges Safety Performance Theory with Socio-Technical Systems Theory, positing that occupational safety outcomes are emergent system properties generated by the dynamic coordination of technical controls and human subsystems rather than isolated administrative protocols. Guided by this systems-thinking perspective, the investigators conducted an extensive scoping process across international guidelines issued by the Occupational Safety and Health Administration, the International Labour Organization, and the American Petroleum Institute, alongside peer-reviewed safety literature and compliance records from the East Azerbaijan Gas Company and the National Iranian Gas Company spanning 2005 to 2024. This initial scoping exercise established a preliminary pool of eighty-four candidate indicators representing diverse occupational, operational, and environmental constructs. To filter and refine this extensive pool, the team convened an expert review panel comprising fourteen subject-matter specialists holding university degrees in related engineering or management disciplines and averaging 10.8 years of professional experience in high-risk safety auditing and contractor oversight.
The refinement and validation of the indicator pool followed a rigorous, multi-stage screening methodology based on psychometric and content validity protocols. In the initial screening phase, expert review eliminated three overlapping items that exhibited semantic redundancy, refining the candidate set to eighty-one measures. The expert panel then evaluated the remaining indicators using a nine-point rating scale across dimensions of operational relevance, definitional clarity, and field implementation feasibility. Applying Lawshe’s methodology, the minimum acceptable Content Validity Ratio was established at 0.51 for the fourteen-member panel, ensuring that retained indicators were formally judged as essential by a clear majority of the participating experts. Following this initial filtering, indicators meeting a Content Validity Index threshold of 0.78 were confirmed for representativeness, systematically condensing the candidate pool into forty-three validated sub-indicators. Psychometric reliability analysis across the final consolidated structure demonstrated exceptional internal consistency, yielding an overall Cronbach’s alpha coefficient of 0.91, with individual dimension values ranging stably between 0.81 and 0.92 across all structural areas.
To accommodate the interdependent feedback loops inherent in contractor safety ecosystems, the researchers organized the forty-three validated sub-indicators into seven principal thematic dimensions: Awareness and Competence, Risk Assessment and Management, Management Systems, Environmental Health Management, Monitoring and Auditing, Emergency Preparedness, and Documentation. Recognizing that conventional linear scoring systems and hierarchical models such as the Analytic Hierarchy Process incorrectly assume unidirectional criteria independence, the investigators adopted the Fuzzy Analytic Network Process implemented in Super Decisions software. This method captured both intra-cluster and inter-cluster reciprocal influences through expert pairwise comparisons recorded on Saaty’s standard one-to-nine scale. Linguistic subjectivity and epistemic uncertainty in expert judgments were accommodated through triangular fuzzy numbers aggregated via the geometric mean method and defuzzified using the centroid technique. Consistency ratios across all comparative evaluation matrices averaged 0.054 and consistently satisfied the strict acceptance threshold of less than 0.10, confirming high internal judgment coherence throughout the network weighting process.
Analysis of the global priority weights derived through the Fuzzy Analytic Network Process revealed that experts placed the highest structural importance on organizational capability and preventative governance. Among the seven primary dimensions, Awareness and Competence achieved the largest global priority weight at 25.8 percent, closely followed by Risk Assessment and Management at 21.9 percent, while WHSE Management Systems secured 14.8 percent and Environmental Health Management captured 21.0 percent. Monitoring and Auditing was allocated a 17.0 percent weighting, whereas Documentation recorded the lowest global priority at 6.7 percent. At the granular sub-indicator level, specific operational measures emerged as dominant priorities within their respective functional clusters, led by ISO Certifications at 25.1 percent, Formal WHSE Training at 24.2 percent, and Risk Identification and Assessment Systems at 21.0 percent. These derived weights demonstrate that safety professionals conceptually prioritize foundational training, systematic risk identification protocols, and structured organizational management architectures when establishing baseline criteria for long-term contractor governance in hazardous environments.
To test the operational stability of the derived weight architecture, the researchers executed a comprehensive sensitivity analysis assessing the effect of positive and negative ten percent perturbations across all seven indicator clusters. For each specific perturbation scenario, modified performance index scores were recalculated across participating contractor profiles while holding all remaining cluster weights completely constant. The empirical results demonstrated remarkable stability across the entire analytical framework, as not a single evaluated contractor shifted across established compliance tiers during the weight perturbation trials. The maximum rank displacement observed in the dataset was restricted to a shift of two positions, which occurred exclusively during perturbations applied to the Documentation cluster. Furthermore, Spearman’s rank correlation coefficients across all examined perturbation scenarios remained consistently above 0.96, quantitatively confirming that the structural rankings generated by the index are exceptionally resilient against minor variations or potential subjectivities in expert weighting inputs.
The practical discriminative capability of the index was subsequently evaluated through case study applications and large-scale descriptive analyses involving operational contractor compliance data from gas infrastructure projects. A detailed case study conducted on five representative contractors operating under strict regulatory oversight classified all five firms into a Moderate compliance tier defined by composite scores ranging between 75.0 and 84.9 points, with individual scores spanning from 75.45 to 81.18 points. Rather than reflecting an inability to categorize performance, the researchers emphasized that the five-point spread and distinct domain profiles revealed meaningful differences in safety maturity that conventional binary or checklist methods entirely obscure. Broadening the empirical investigation to descriptive evaluations across two hundred active contractors, overall composite scores ranged from 48.5 to 87.2 with an overall mean score of 67.87 and a standard deviation of 9.69. Among individual operational dimensions, Environmental Health recorded the highest mean score at 79.01, whereas Risk Management exhibited the lowest mean at 61.16, with Documentation and Awareness displaying the widest dispersion across contractor organizations.
To independently cross-validate the expert-derived decision network against real operational outcomes, the authors trained a Random Forest regression model using standardized compliance records from twenty contractors with complete archival datasets. Operating with five hundred regression trees, bootstrap sampling, and mean squared error split criteria, the machine learning model was validated using a ten-fold cross-validation procedure to mitigate small-sample overfitting risks. The predictive model achieved a coefficient of determination of 0.62, an average Mean Absolute Error of 1.41, and a Mean Squared Error of 3.33, demonstrating robust predictive performance and substantial explanatory power without evidence of destructive overfitting. An analysis of actual versus predicted performance scores along the identity line confirmed strong alignment across mid-to-high performance ranges. Crucially, the model was positioned as an analytical triangulation mechanism designed to assess empirical indicator relevance rather than serve as an autonomous, generalizable predictive engine across external operational settings.
Comparing the empirical feature importance values from the Random Forest model with the expert-derived weights from the network process revealed striking convergences as well as operational divergences. While Environmental Health Management exhibited near-perfect agreement between the machine learning importance of 22.5 percent and its network weight of 21.0 percent, substantial divergence emerged in Monitoring and Auditing, which captured the highest predictive importance in the machine learning model at 37.6 percent despite receiving only 17.0 percent in expert weighting. Conversely, Awareness and Competence, which held the primary position in expert weighting, demonstrated substantially lower direct importance in predictive modeling. The authors explained that while safety experts view training and competence as essential cultural foundations, their benefits manifest over extended time horizons, whereas active monitoring, site inspections, and environmental safeguards exert immediate, measurable impacts on short-term compliance scores and operational incident reduction across active engineering projects.
The study authors highlighted several methodological considerations, noting that the expert panel was limited to fourteen domain specialists from regional operations and that machine learning validation was restricted to a standardized subset of twenty contractors due to archival documentation completeness. Nevertheless, the investigation substantiates that integrating network-based decision modeling with empirical machine learning produces an adaptable, objective decision-support instrument that bridges theoretical safety science and field-level operational governance. By moving beyond traditional lagging indicators and unweighted compliance checklists, the performance index provides industrial asset owners and regulatory authorities with a transparent mechanism for pre-contract contractor screening, periodic site auditing, and targeted risk interventions. The authors concluded that future iterations of the framework could implement dynamic hybrid weighting schemes that blend baseline structural priorities with responsive machine learning adjustments, supporting performance-based safety regulation across petrochemical, mining, construction, and related high-hazard sectors.
Subject of Research: Contractor health, safety, environment, and quality performance evaluation in the gas industry
Article Title: A hybrid ANP and machine learning framework for contractor health safety environment and quality performance evaluation
Article References: Khordoustan, P., Akbarzadeh, O., Gilani, N., Moshashaei, P., & Alizadeh, S. S. (2026). A hybrid ANP and machine learning framework for contractor health safety environment and quality performance evaluation. Discover Analytics, 4(1), Article 16. https://doi.org/10.1007/s44257-026-00060-x
Image Credits: AI Generated
DOI: 10.1007/s44257-026-00060-x
Keywords: contractor safety, fuzzy ANP, machine learning, occupational safety, oil and gas, risk assessment, safety governance, performance indicators, hybrid, machine, learning, framework
News Source: Denise Maddox. (November 13, 2026). Hybrid Analytical Framework Enhances Contractor Safety Performance Evaluation in Gas Operations. Scienmag.
