The Dempster-Shafer data-fusion technique as affected by probability masses as a result of sensor selection and probability masses distribution is investigated. Dempster-Shafer inference is a statistically based data-classification technique for detecting traffic events that affect normal traffic operations. It is used when data sources contribute discontinuous and incomplete information such that no single data source can produce a predominantly high probability of certainty for identifying the most probable event. To help in the selection of appropriate sensors and probability masses, a rough-sets data-mining technique in support of Dempster-Shafer inference was proposed and tested. The basic rough-sets technique is introduced, and a numerical example is given to explain its applications. Field testing of the rough-sets technique showed that it can reasonably and systematically process a large amount of traffic information, as an alternative to relying on the intuition of traffic operators and system managers. Because it allows easy maintenance and update of estimated probability masses, this technique is suitable for large-scale applications at the traffic management center.
Abstract