Real-Time Framework to Predict Crash Likelihood and Cluster Crash Severity

Abstract

This study proposes a three-stage framework for real-time crash likelihood and severity prediction. Firstly, a real-time crash likelihood prediction model was developed. Secondly, a real-time crash severity clustering model was proposed to cluster the crashes into different severity levels. Thirdly, a severity clustering validation model was developed to assess the performance of the proposed severity clustering model. Extensive data processing techniques were employed to collect real-time features from State Road 408 in Orlando, Florida, and a total of 6,750,072 events (625 crash events and 6,749,447 non-crash events) along with 24 real-time features were used. To develop the crash likelihood prediction model, nine machine-learning techniques were attempted, and the convolutional neural network model was found to provide the best result with respect to the sensitivity (0.916), false alarm rate (0.111), and area under the receiver operating characteristic curve (0.967). Davies–Bouldin index criteria were used to find the detector location that generated the most accurate traffic information to cluster the crashes into severity levels, and based on this traffic information, k-means clustering was applied to develop the severity clustering model. Finally, a severity clustering validation model was developed after investigating nine machine-learning techniques to validate the developed severity clustering model, and the decision tree model provided the best results based on three levels of sensitivity and specificity values. The developed framework has the potential to help traffic management centers to warn road users or develop transportation systems management and operations strategies in real time to avoid crashes or minimize the severity and, thus, can significantly contribute to improving road safety.

Calibrated confidence learning for large-scale real-time crash and severity prediction

“…Some efforts to assess the viability of spatial and temporal transferability of real-time crash likelihood prediction models 30 or propose a workflow for real-time crash likelihood prediction, quantification, and classification 39 were also found. Real-time crash severity-related studies have evolved in the literature since 2010, and these studies are conducted in the USA 11,[40][41][42][43][44][45][46][47][48][49] , China 50,51 , France 52 , Greece 53 , and Iran 54 . Most of these studies are conducted on Freeways/Expressways/Toll Roads.…”

“…The length of the roadway section considered in those studies ranged from very low, i.e., 7.7 miles 49 , to sufficiently large, i.e., 111 miles 47 . The real-time data study period used for modeling or analyses in those studies ranged from 1 year 40,43,46 to 5 years 53,54 .…”

“…Crash data were collected from police crash reports 41 , government organizations, i.e., different departments of transportation (DOTs) 42 , and private entities, i.e., signal four analytics (S4A) 48 . Traffic data were collected mainly from loop detectors 52 , automatic vehicle identification (AVI) 44 , regional integrated transportation information system (RITIS) 47 , automated traffic signal performance measure (ATSPM) 48 , Camera 54 , and microwave vehicle detection system (MVDS) 40 . Most of these traffic data sources update traffic information, i.e., speed, volume and lane occupancy, in every 30 s 43 .…”

“…A few studies also used matched case-control to select non-crash samples, and they used a 1:2 ratio 53 of crash and non-crash samples. Some studies also used the Synthetic Minority Over-sampling (SMOTE) technique to address crash and non-crash data imbalance issues 40 . Then, a dataset is prepared by integrating crash, traffic, weather, road geometric, and other data sources, and further cleaning and processing actions are performed to remove unrealistic and missing values.…”

“…Some used KA, BC, and O combination 41,43,51,52 , whereas some used KAB, C and O combination 42 . Some studies did not follow any global crash-severity scale, rather, they followed their database-provided scale 50 or proposed new crashseverity levels based on the crash severity likelihood estimated based on traffic state and traffic-flow diagram 40 . Such studies used three levels of crashseverity, i.e., High/Severe, Medium and low/light.…”