Fuzzy Data Mining Approach for Quantifying Signalized Intersection Level of Services Based on User Perceptions

This paper deals with how users perceive quality of service at signalized intersections and how many levels of service (LOS) drivers are able to perceive. A carefully designed laboratory experiment was conducted to create a user perception database of 100 subjects assessing 24 approaches in terms of stopped delay and rating of service. The data were subsequently processed to detect abnormal records and unusual subjects before any data mining tool was applied. The preprocessed data were then mined for hidden knowledge using a fuzzy $c$ -means data clustering technique, which is unsupervised learning for discovering distinct clusters of user perceived delays and rating of services. The clustering results are presented and analyzed according to approach membership, delay membership, and rating membership of the clusters. The conclusion is that drivers are able to differentiate among six levels of service, but not the existing Highway Capacity Manual (HCM) ones. A set of new six levels of service are proposed, with the existing HCM LOSs A and B merged for a single level and the existing HCM LOS F split into two. The study demonstrates that fuzzy membership is appropriate to address subjective perception, and with the aid of fuzzy membership, each new level of service has its primary members of full degree and secondary members of varying degrees.