Abstract:
To prevent the occurrences of accidents at workplaces, accident data should be analyzed properly. However, handling such data of higher dimension is often a difficult task for analysis to achieve efficient decision making due to the slow convergence and local minima problem. To address these issues, the present study proposes a new clustering algorithm called growing self-organizing map (GSOM)-based genetic K-means (GSGKM) for classifying accident data into an optimal number of clusters. Tolerance rough set approach (TRSA) is later used on each cluster to extract useful accident patterns, which enables helps in accident analysis and prevention. To validate the effectiveness of our proposed methodology, accident data obtained from an integrated steel plant are used as a case study. Besides, a total of four benchmark datasets collected from the University of California, Irvine (UCI) machine learning repository are also used for comparative study to prove its (i.e., GSGKM) superiority over some other state-of-the-arts. Experimental results reveal that the proposed methodology provides the highest clustering accuracy. A total of four clusters are obtained from the analysis. A set of 16 accident crisp patterns or rules are extracted from clusters using TRSA. Company employees are found to be more exposed to accidents than contractors. Additionally, behavioral issues are identified as the most determinant factor behind the injuries at work. The proposed methodology can be effectively used in decision making for different industries, including construction, manufacturing, and aviation.