EXPLORING VISIBILITY RELATED CRASHES ON FREEWAYS BASED ON REAL-TIME TRAFFIC FLOW DATA

摘要

There is a lack of prior studies that investigated the relationship between traffic flowvariables and traffic crashes that occur due to reduced visibility. This paper aims at exploring theoccurrence of visibility related (VR) crashes on freeways using real-time traffic surveillance data(speed, volume and occupancy) collected from underground loop detectors (LD) and radarsensors potentially associated with VR crash occurrence. The research hypothesis here is tocompare traffic flow characteristics leading to VR crashes with non-crash cases at reducedvisibility conditions. Historical crash and LD data were collected from Interstates 4 and 95 inFlorida between December 2007 and March 2009.To achieve the objectives of this study, Random Forests (RF), a relatively recent datamining technique, was used to indentify significant traffic flow variables affecting VR crashoccurrence. Using significant variables selected by RF, matched case-control logistic regressionmodel was estimated. The purpose of using this statistical approach is to explore the effects oftraffic flow variables on VR crashes while controlling for the effect of other confoundingvariables such as the geometric design elements of freeway sections (i.e. horizontal and verticalalignments) and crash time.The results revealed that the 5-minutes average occupancy observed at the nearestdownstream station during 10-15 minutes before the crash along with the average speedmeasured at the downstream and upstream stations during 5-10 minutes before the crash increasethe likelihood of VR crash occurrence in between. In addition, by using a threshold value of 1.0for the corresponding odds ratio, over 67% VR crash identification was achieved.