A Parallel FP-Growth Mining Algorithm with Load Balancing Constraints for Traffic Crash Data

Traffic safety is an important part of the roadway in sustainable development. Freeway traffic crashes typically cause serious casualties and property losses, being a serious threat to public safety. Figuring out the potential correlation between various risk factors and revealing their coupling mechanisms are of effective ways to explore and identity freeway crash causes. However, the existing association rule mining algorithms still have some limitations in both efficiency and accuracy. Based on this consideration, using the freeway traffic crash data obtained from WDOT (Washington Department of Transportation), this research constructed a multi-dimensional multilevel system for traffic crash analysis. Considering the load balancing, the FP-Growth (Frequent Pattern-Growth) algorithm was optimized parallelly based on Hadoop platform, to achieve an efficient and accurate association rule mining calculation for massive amounts of traffic crash data; then, according to the results of the coupling mechanism among the crash precursors, the causes of freeway traffic crashes were identified and revealed. The results show that the parallel FPgrowth algorithm with load balancing constraints has a better operating speed than both the conventional FP-growth algorithm and parallel FP-growth algorithm towards processing big data. This improved algorithm makes full use of Hadoop cluster resources and is more suitable for large traffic crash data sets mining while retaining the original advantages of conventional association rule mining algorithm.