Discovering Indirect Associations in Crash Data through Probe Attributes

Abstract

Association analysis is a popular data-mining technique for detecting dependencies in transaction databases. The algorithms for discovery of association rules attempt to find the products that tend to sell together. For this purpose, one can treat crashes as individual transactions and crash characteristics as the “products” part of that transaction to discover patterns of the crash characteristics that tend to coexist. The a priori algorithm to search for association rules employs the lower bound on intersection (i.e., joint) frequency of two crash characteristics to ensure that only significant patterns are discovered. Hence, remarkable associations relating crash characteristics that have joint frequency lower than the specified threshold cannot be identified. The present study proposes to apply probe attributes and indirect associations to overcome this problem for crash data. Two sets of items, X and Y, are indirectly related if there exists another set P (i.e., probe attribute) such that, while sets X and Y rarely or never occur simultaneously, they are highly dependent on the common probe attribute P. Indirect associations may be particularly useful in the context of arterial crash patterns because several products are part of the same original nominal variable and have no intersection (i.e., zero joint frequency) in the market basket database. For example, the products rain and cloudy created from the same original variable weather have zero joint frequency. The probe attributes-based analysis revealed some remarkable arterial crash patterns for the SR-50 corridor (in central Florida), including an association between morning and afternoon peak-period crashes through the site location driveway access. It was also found that young drivers are prone to commit errors on arterial sections that are not level, straight, or both.

Get full access to this article

View all access options for this article.

References

Safety Data Action Plan. Bureau of Transportation Statistics, Department of Transportation, Sept. 13, 2000. http://www.bts.gov/publications/safety_data_action_plan/entire.pdf. Accessed May 2007.

Pande

and Abdel-Aty

M. A.

Market Basket Analysis: Novel Way to Find Patterns in Crash Data from Large Jurisdictions.

Presented at 86th Annual Meeting of the Transportation Research Board, Washington, D.C., 2007.

Geurts

Thomas

and Wets

Understanding Spatial Concentrations of Road Accidents Using Frequent Item Sets. Accident Analysis and Prevention, Vol. 37, No. 4 2005, pp. 787–799.

Agrawal

Imielinski

and Swami

Mining Association Rules Between Sets of Items in Large Databases. Proc., ACM SIGMOD International Conference on Management of Data, Washington, D.C., May 25–28, 1993, pp. 207–216.

Getting Started with Enterprise Miner Software, Release 4.1. SAS Institute, Cary, N.C., 2001.

Hand

Mannila

and Smyth

Principles of Data Mining. MIT Press, Cambridge, Mass., 2001.

Brin

Motwani

and Silverstein

Beyond Market Baskets: Generalizing Association Rules to Dependence Rules. Data Mining and Knowledge Discovery, Vol. 2, No. 1 1998, pp. 39–68.

Das

Mannila

and Ronkainen

Similarity of Attributes by External Probes. Proc., KDD'98: 4th International Conference on Knowledge Discovery and Data Mining, New York, Aug. 27–31, 1998.

Tan

P. N.

Kumar

and Kuno

Using SAS for Mining Indirect Associations in Data. Proc., 9th Western Users of SAS Software Conference, San Francisco, Calif., Sept. 5–7, 2001.

10.

Tan

P. N.

Kumar

and Srivastava

Indirect Association: Mining Higher Order Dependencies in Data. Proc., 4th European Conference on Principles and Practice of Knowledge Discovery in Databases, Lyon, France, Sept. 13–16, 2000, pp. 632–637.