Mining of inland water traffic accident data using a biclustering algorithm: A case study of the Yangtze River

Analysis of maritime accident data is important for improving safety management. Clustering is the favoured method of mining marine accident data. However, traditional one-way clustering methods are limited by their focus on global patterns, which does not account for the contingent characteristics of accidents. In this study, biclustering algorithms (BAs) typically used for gene expressions are introduced for analysis of inland water traffic accident data. BAs are good for discovering local patterns (LPs), which represent the similarities between partial accidents and partial attributes. LPs are the more likely modes in accident data, which are difficult to discern using who is traditional one-way clustering. During biclustering of original accident data, six LPs involving replicative accidents are uncovered, thereby suggesting a high risk in similar scenarios. With biclustering of accident attribute factors, the interrelationships among factors are discovered. According to the LPs explored using BAs, high-risk scenarios should gain the attention of shipping companies and safety management departments. Two recommendations are presented: raising awareness of the need for immediate accident reporting and disseminating rescue knowledge. After comparing their applications, the order-preserving submatrix (OPSM) and conserved gene expression motif (xMotifs) algorithms are regarded as the most suitable BAs for analysing maritime accident data