Application of data mining techniques for real-time crash risk assessment on freeways

Data mining is the analysis of large "observational" datasets to find unsuspected relationships that might be useful to the data owner. It typically involves analysis where objectives of the mining exercise have no bearing on the data collection strategy. Freeway traffic surveillance data collected through underground loop detectors is one such "observational" database maintained for various ITS (Intelligent Transportation Systems) applications such as travel time prediction etc. In this research data mining process is used to relate this surrogate measure of traffic conditions with rear-end crash 36.25-mile occurrence on freeways. Crash and dual loop detector data from S instrumented Interstate-4 corridor in Orlando (FL) are used in this study. The research problem is set up as a classification problem and separate data mining based classifiers are developed to discriminate crashes belonging to different categories from normal conditions on the freeway. Based on the models developed in this study one can identify the traffic conditions prone to rear-end crashes 5-10 minutes prior to the crash. The findings of this research are proposed to be used as a proactive traffic management system which could warn the drivers about potential rear-end crashes. Introduction The objective of this research is development of a framework to detect crash prone conditions in real-time. To achieve these objectives loop data collected from randomly selected non-crash locations have been used in this study along with the crash data. These data most commonly include speed, vehicle counts, and lane occupancy provided every 30 seconds by loop detectors installed beneath the freeway pavement. To establish relationships between real-time traffic data, geometric parameters, and rear-end crashes a data mining approach is adopted. It essentially means that tools from a range of fields such as machine learning (e.g., clustering algorithms), statistics (e.g., classification tree), and/or artificial intelligence are used in a step by step manner to analyze the data. This research is part of a new trend in freeway traffic management which until recently was focused on timely detection of incidents. With the enormous increase in mobile phone usage in the recent past relevance of incident detection is diminishing and traffic management authorities are looking for proactive strategies. The basic element of a proactive traffic management system would be reliable models separating crash prone conditions from 'normal' traffic conditions in real-time. Most of the existing real-time crash 'prediction' models available in the literature are generic in nature, i.e., single generic model has been used to identify all crashes (such as rear-end, sideswipe, or angle). Conditions preceding crashes are likely to differ by type of crash and therefore the approach towards proactive traffic management should be type (of crash) specific in nature. The disaggregate models would also be useful in devising specific countermeasures for crashes. In this research the focus is on the most frequent group of crashes on the freeways, L the rear-end crashes. The rear-end crash data for this study are collected over a five year period (1999 through 2003) from S corridor of Interstate-4 in Orlando metropolitan area along with information about geometric design features, such as ramp locations, curvature, etc. The corridor has a total of 69 loop detector stations in each direction, spaced out at nearly half a mile. Each of these stations consists of dual loops and measures average speed, occupancy, and volume over 30-second period on each of the three through travel lanes in both directions.