Interpretable Safety Priority Ranking for Low-Rise Construction Sites Using a PCA–ANN–Garson Framework
Springer Series in Geomechanics and Geoengineering, (2026), pp. 432-441
Junjun H. Moreno
a
,
Dante L. Silva
b
,
Kevin Lawrence M. De Jesus
a
a Department of Civil Engineering, FEU Institute of Technology, Sampaloc, Manila, 1015, Metro Manila, Philippines
b School of Graduate Studies, Mapúa University, Intramuros, Manila, 1002, Metro Manila, Philippines
Abstract: Low-rise (1–4 storey) building projects often operate under time and re-source constraints that can weaken the consistency of safety implementation. This study proposes an interpretable, data-driven framework that converts multi-indicator safety assessments into a defensible priority ranking for site interventions. A content-validity–tested questionnaire (24 indicators) was evaluated by eight construction experts and administered to Philippine site engineers and safety officers, yielding 260 valid responses. The dataset satisfied Principal Component Analysis (PCA) prerequisites (Kaiser–Meyer–Olkin (KMO) = 0.7791; Bartlett’s χ2 = 4617.4677, df = 325, p < 0.05). Seven principal components were retained (eigenvalue >1), ex-plaining 79.006% of total variance and forming orthogonal component-level safety dimensions. These component scores were used as inputs to an Artificial Neural Network (ANN) model trained with Levenberg–Marquardt; across 30 random initializations, the most stable architecture was Hidden Neurons (HN) = 1 (validation: Mean Absolute Percentage Error (MAPE) = 6.297% ± 0.487; R = 0.82). For benchmarking, linear and ridge regression achieved validation MAPE of 6.5039% (R = 0.81338) and 6.3908% (R = 0.81185), respectively. To enable interpretability, Garson’s algorithm computed relative importance across seeds, showing a dominant general safety factor (PC1: 64.887% ± 6.962) followed by domain-specific drivers (PC2: 10.648% ± 2.1105; PC5: 8.0125% ± 3.6733; PC6: 7.3157% ± 3.2195). The resulting PCA–ANN–Garson workflow supports evidence-based prioritization of safety improvements in low-rise construction sites by combining predictive accuracy with transparent ranking. This work supports Sustainable Development Goal (SDG) 3 (Good Health and Well-being) and SDG 8 (Decent Work and Economic Growth) through improved prevention of site injuries and stronger working-condition safeguards.