Assessment of freeway traffic parameters leading to lane-change related collisions

Abstract

This study aims at 'predicting' the occurrence of lane-change related freeway crashes using the traffic surveillance data collected from a pair of dual loop detectors. The approach adopted here involves developing classification models using the historical crash data and corresponding information on real-time traffic parameters obtained from loop detectors. The historical crash and loop detector data to calibrate the neural network models (corresponding to crash and non-crash cases to set up a binary classification problem) were collected from the Interstate-4 corridor in Orlando (FL) metropolitan area. Through a careful examination of crash data, it was concluded that all sideswipe collisions and the angle crashes that occur on the inner lanes (left most and center lanes) of the freeway may be attributed to lane-changing maneuvers. These crashes are referred to as lane-change related crashes in this study. The factors explored as independent variables include the parameters formulated to capture the overall measure of lane-changing and between-lane variations of speed, volume and occupancy at the station located upstream of crash locations. Classification tree based variable selection procedure showed that average speeds upstream and downstream of crash location, difference in occupancy on adjacent lanes and standard deviation of volume and speed downstream of the crash location were found to be significantly associated with the binary variable (crash versus non-crash). The classification models based on data mining approach achieved satisfactory classification accuracy over the validation dataset. The results indicate that these models may be applied for identifying real-time traffic conditions prone to lane-change related crashes.

A Data Mining Approach to Identify Key Factors of Traffic Injury Severity

“…The VIM for this variable is the weighted average of the reduction in the Gini impurity measure achieved by all splits using variable x j across all internal nodes of the tree, and the weight is the node size. If N is the total number of observations in the training sample, then the formula for the importance for variable x j is given by the following [15]:…”

“…Analyzing 2001 Florida crash database, they found that rear-end crashes are over-presented in the higher speed limits (45-55mph), and the rearend crash propensity is larger for daytime, wet and slippery road surface conditions, male gender, and drivers younger than 21 years of age. Pande and Abdel-Aty [15] in another study employed classification tree based variable selection procedure to identify the important parameters leading to lane-change related freeway crashes, using the traffic surveillance data collected from a pair of dual loop detectors.…”

Traffic Accident Data Profiling and Clusteringwith Data Mining Process

“…There is a dramatic reduction in the rate of road crash from 1988. Analysts have ascribed this improvement to the activities of the Federal Road Safety Commission in Nigeria since 18 th February 1988. The mandate of the Federal Road Safety Commission is to reduce the rate of traffic crashes to the minimum since at this moment, zero vision has not been contemplated in Nigeria.…”

Real-time crash prediction on freeways using data mining and emerging techniques

“…Even though, some studies [14] find that a generalized linear model-based approach may lead to biased estimates when the independent variables demonstrate strong nonlinear features. The commonly used techniques in predicting real-time crash likelihood are Neural Network (NN) [3,15,16] and Support vector machines (SVMs) [17,18].…”