Dynamic Risk Assessment of Tower Crane Operations Using DEMATEL-ISM and a Dynamic Bayesian Network

Continuous tower crane monitoring enables the recording of changing operating conditions, but interpreting these measurements requires a model of how operational risk states evolve. We combine the Decision-Making Trial and Evaluation Laboratory (DEMATEL), interpretive structural modeling (ISM), and a two-slice dynamic Bayesian network (DBN) for dynamic risk assessment. Six operating indicators define a relative operational risk state from training-referenced percentiles in 60 s time slices. Coverage and activity screening identify eligible observations, while lagged associations and DEMATEL-ISM guide DBN structure screening. The DBN predicts the next-slice state probability and relates each forecast to its inputs through conditional probability contrasts. The case study used one month of records from a tower crane, divided chronologically into training, validation, and test sets. Across 2557 test pairs, the model achieved an area under the receiver operating characteristic curve of 0.6931 and average precision of 0.4571. Its Brier score was 0.1724, compared with 0.1899 for a Markov baseline using only the current risk state. This Brier advantage persisted across eight state definitions, and coupling trend consistently had the largest mean conditional probability contrast. By connecting operating measurements with temporal prediction and factor interpretation, the method supports systematic tracking of changes in tower crane operational risk states.