Application of Data Fusion via Canonical Polyadic Decomposition in Risk Assessment of Musculoskeletal Disorders in Construction: Procedure and Stability Evaluation

Abstract

Missing data is a common problem in data collection for work-related musculoskeletal disorder (WMSD) risk-assessment studies. It can cause incompleteness of risk indicators, leading to erroneous conclusion on potential risk factors. Previous studies suggested that data fusion is a potential way to solve this issue. This research evaluated the numerical stability of a data fusion technique that applies canonical polyadic decomposition (CPD) for WMSD risk assessment in construction. Two knee WMSD risk-related data sets—three-dimensional (3D) knee rotation (kinematics) and electromyography (EMG) of five knee postural muscles—collected from previous studies were fused for the evaluation. By comparing the consistency performance with and without data fusion, it revealed that for all low to high proportion of missing data (10%–70%) from both kinematics and EMG data sets, the WMSD risk assessment using fused data sets outperformed using unfused kinematics data sets. For large proportions of missing data (>50%) from both kinematics and EMG data sets, better performance was observed by using fused data sets in comparison with unfused EMG data sets. These findings suggest that data fusion using CPD generates a more reliable risk assessment compared with data sets with missing values and therefore is an effective approach for remedying missing data in WMSD risk evaluation.