1 / 43

Freeway Crash Prediction Models for Long-Range Urban Transportation Planning

Freeway Crash Prediction Models for Long-Range Urban Transportation Planning. Arun Chatterjee (UT) Joe Hummer (NCSU) Vasin Kiattikomol (UT) Mary Sue Younger (UT). Objective.

hashim
Download Presentation

Freeway Crash Prediction Models for Long-Range Urban Transportation Planning

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. Freeway Crash Prediction Models for Long-Range Urban Transportation Planning Arun Chatterjee (UT) Joe Hummer (NCSU) Vasin Kiattikomol (UT) Mary Sue Younger (UT)

  2. Objective • Develop crash prediction models to assess safety impacts for urban freeway networks, which can be used by state DOT and MPO planners for long-range planning • The independent variables should not be too difficult to predict

  3. How to Create Freeway Segments? • No previous work recognized the influence of interchanges on crashes explicitly

  4. Impact of Interchanges on Crashes • Since traffic flow characteristics near interchanges are different from areas away from them, interchanges are likely to have impact on crashes • No previous studies examined this issue • This research compared crashes in interchange areas with those in areas away from it, and found a significant difference

  5. Non-interchange segments Interchange segments 1,500’ 1,500’ Definition of Segments Used in this Research

  6. Crash Rates for NC & TN Unit: crashes per Million Vehicle Miles Traveled

  7. Crash Rates for NC & TN Unit: crashes per Million Vehicle Miles Traveled 1 2000-2002 statewide crash report, NCDOT

  8. Proportion of Crash Types 12000-2002 statewide crash report, NCDOT 21988-2003 crash data, National Highway Traffic Safety Admin.

  9. Model Development Scheme • Develop separate models for: • Segments away from interchanges • Segments with 4 lanes • Segments with more than 4 lanes • Segments near interchanges • Segments with 4 lanes • Segments with more than 4 lanes • 3 Modeling Approaches • Analysis of Variance (ANOVA) • Regression Analysis • Classification Tree Analysis

  10. Crash and Freeway Data • Counties with medium to large sized urban areas • NC: 6 counties • TN: 4 counties • Freeway data • NC: 73 non-interchange segments (137 mi.), 65 interchange segments • TN: 142 non-interchange segments (145 mi.), 65 interchange segments • Crash data by severity (2000-2002) • Fatal • Injury • Property-damage-only (PDO)

  11. Counties • NC Counties: • Alamance, Buncombe, Cumberland, Durham, Gaston, and Wake • TN Counties: • Davidson (Nashville), Hamilton (Chattanooga), Knox (Knoxville), Shelby (Memphis)

  12. Model Classification

  13. Crash Prediction Models • NC • TN • Non-interchange segments • Interchange segments Model classification for each modeling approach: • By state • By type of segments • By number of lanes • By type of crash severity • 4 lanes • >4 lanes • Injury • PDO • Fatal + injury

  14. ANOVA Models • Response variable: Crash rate (by severity) • Factors: • Type of segments • Number of lanes • Traffic volume • Number of models • 3 for NC • 3 for TN

  15. Example: Fatal and Injury Crash Rates for NC

  16. Total Crash Rates for NC Total = Fatal + Injury + PDO

  17. Example: PDO Crash Rates for TN

  18. Total Crash Rates for TN Total = Fatal + Injury + PDO

  19. Regression Models • Response: No. of Crashes (by severity) • Predictors: • Segment length • Traffic volume (AADT) • Use Generalized Linear Modeling technique with Negative Binomial Distribution assumption for crashes • Estimate model parameters using maximum likelihood technique in ‘GENMOD’ procedure in SAS

  20. Regression Models (Contd.) • Common model forms: • Crashes = a (Segment Length)b1 (AADT)b2 for non-interchange segments • Crashes = a (AADT)b for interchange segments • Number of models • 12 for NC • 12 for TN

  21. Example: Models for Non-Interchange Segments of NC Note: R2 is special type (Rk2)

  22. Example: Models for Interchange Segments of NC Note: R2 is special type (Rk2)

  23. Predicted Fatal and Injury Crashes for NC Models

  24. Predicted PDO Crashes for NC Models

  25. Predicted Total Crashes for NC Models

  26. Predicted Fatal and Injury Crashes for TN Models

  27. Predicted PDO Crashes for TN Models

  28. Predicted Total Crashes for TN Models

  29. Findings of Graphical Analysis • For NC, ‘interchange crashes’ on 4-lane freeways are higher and also increased more rapidly at higher volumes than others • For TN, ‘interchange crashes’ on >4-lane freeways are higher at higher volumes than others • The pattern of increase of ‘fatal and injury’ crashes is similar to that of ‘PDO’ crashes in both NC and TN

  30. Classification Tree Models • Predict: Crash Rate (by severity) • Predictors: • Traffic volume (AADT) • Number of Lanes • Develop in 2 steps: • Develop tree models using CART algorithm in SPSS AnswerTree • Manually calculate crash rate using VMT information and predicted crashes from tree models and develop crash rate tables

  31. Classification Tree Models (Contd.) • Number of models • 2 for NC • 2 for TN

  32. Classification Tree Model for Interchange Segments of TN

  33. Crash Rate Table for Interchange Segments of TN

  34. Classification Tree Model for Interchange Segments of NC

  35. Crash Rate Table for Interchange Segments of NC

  36. Findings for Classification Tree Models • Each tree has unique structure (split pattern) for each state and for each type of segments • Tree models classified segments into groups with different threshold values of AADT from those specified in ANOVA models

  37. Assessment and Comparison of Models • Assess model performance using ‘relative measure’ in terms of ‘percent error’, which is calculated in 2 steps: • Calculate difference between predicted crashes and actual crashes in terms of Root Mean Squared Error (RMSE) for each model • Calculate Percent Error of each model

  38. Assessment and Comparison for Models for NC

  39. Assessment and Comparison for Models for TN

  40. Conclusions • Crash rates for non-interchange segments and those for interchange segments are different • Developing separate models for non-interchange segments and interchange segments is practical • Crash prediction models developed based on different modeling approaches performed fairly similar except for • Non-interchange segments with >4 lanes for NC • Interchange segments with 4 lanes for TN

  41. Future Directions • Develop models for urban areas in other states • Develop models for other types of roadway facilities such as expressways • Use classification tree technique to fine-tune the grouping of traffic volume in ANOVA models

  42. THANKS • THANKS FOR YOUR HELP

More Related