A Novel Approach for Signalized Intersection Crash Classification and Prediction

The authors describe a study that impacts signalized intersection safety by identifying geometric characteristics that affect different crash types. In the first part, crash frequency was estimated using intersection traffic volume and geometric volume. Averaging the number of crashes by total number of approach lanes then by crash frequency for each intersection type provided an estimate that allowed intersections to be classed as "safe" or "unsafe." The model then allowed a simulation database with all possible geometric and traffic volume intersection combinations to estimate crash frequency, the result of which was used to study how intersection properties affected crash frequency. Most significant in determining intersection safety were right turn channelization, number of left turning lanes, and number of through lanes on minor roadways. In the second part of the study, developing the "Neural Network Trees" innovative approach allowed intersection traffic volume, geometric factors, and conditions at time of crash to be used to classify signalized intersection crashes as rear-end, angle, turn and sideswipe. The first neural network model built into the "Trees" classified both same direction (sideswipe and rear end) and intersecting (turn and angle) crashes. The next neural network models classified individual crash types according to such factors as ADT, number of lanes (through and left turn), right turn channelization, and speed limits. Greater insight into crash type occurences was gained through using models on a simulation database. Promising safety research techniques, including Multi Layer Perceptron (MLP), Probabilistic Neural Networks (PNN), and Generalized Regression Neural Networks (GRNN), were investigated and used in this study. The authors discuss the study conclusion that estimating traffic crashes by safety level or type is superior to estimating one overall aggregate model.

• Availability:
  • Find a library where document is available. Order URL: http://worldcat.org/issn/18245463
• Authors:
  • Abdel-Aty, Mohamed
  • Nawathe, P
• Publication Date: 2006-7

Language

• English

Media Info

• Media Type: Print
• Features: Figures; References; Tables;
• Pagination: pp 67-80
• Serial:
  • Advances in Transportation Studies
  • Volume: 9
  • Publisher: University Roma Tre
  • ISSN: 1824-5463
  • Serial URL: http://www.atsinternationaljournal.com/

Subject/Index Terms

• TRT Terms: Channelized intersections; Classification; Crash rates; Crash types; Crashes; Databases; Estimating; Geometric design; Innovation; Left turns; Mathematical models; Mathematical prediction; Neural networks; Rear end crashes; Right turns; Safety; Side crashes; Signalized intersections; Simulation; Speed limits; Traffic volume; Trees (Mathematics); Turning lanes
• Uncontrolled Terms: Angle collisions; Through lanes; Turn collisions
• Subject Areas: Data and Information Technology; Design; Highways; Safety and Human Factors; I20: Design and Planning of Transport Infrastructure; I82: Accidents and Transport Infrastructure;

Filing Info

• Accession Number: 01042789
• Record Type: Publication
• Files: TRIS, ATRI
• Created Date: Mar 1 2007 10:15AM