1 Introduction
2 Data resources and characteristics
2.1 Different traffic incident time phases
2.2 Data size
2.3 Incident types
2.4 Duration time distribution
2.5 Available data resources
2.6 Significant influencing factors
Types of Factors | Factors |
---|---|
Incident characteristics | Incident severity, incident type, towing requirements, type of involved vehicles, number of casualties, number of lanes blocked and incident location |
Environmental conditions | Rain, snow, dry, or wet |
Temporal factors | Time of day, day of week, season, month of year |
Roadway geometry | Street, intersection, road layout, horizontal/vertical alignment, bottlenecks, roadway type |
Traffic flow conditions | Flow, speed, occupancy, queue length |
Operational factors | Lane closures, freeway courtesy service characteristics |
Vehicle characteristics | Large trucks, trucks with trailers, taxis, special vehicles, compact trucks, number of vehicles involved |
Others | Driver, special events, time that a police officer reaches the site, police response time, report mechanism, accident characteristics reported at accident notification |
2.7 Unobserved heterogeneity and randomness
3 Traffic incident duration analysis
Method Category | Methodology | Researcher | Data source | Duration time phase | Duration distribution |
---|---|---|---|---|---|
Hazard-based duration model (HBDM) | AFT hazard-based model | Jones et al. [41] | 2156 accidents | Response time + clearance time | Log-logistic |
Nam, Mannering [9] | 681 incidents | Detection/reporting, Response time, and Clearance time | Weibull, Weibull, and Log-logistic | ||
Chung et al. [63] | 2940 accidents | Incident duration | Log-logistic | ||
Alkaabi et al. [25] | 583 accidents | Clearance time | Weibull | ||
Chung, Yoon [21] | 1815 accidents | Incident duration | Log-normal | ||
Ghosh et al. [24] | 32,574 incidents | Clearance time | Generalized F | ||
Kaabi et al. [28] | 504 accidents | Response time | Weibull with frailty | ||
Hojati et al. [37] | 4926 incidents | Duration time | Weibulla | ||
Wang et al. [42] | 1198 incidents | Incident duration time | Log-logistic | ||
Chimba et al. [39] | 10,187 incidents | Incident duration time | Log-logistic | ||
Hojati et al. [23] | 430 incidents | Incident duration timeb | Weibull and log-logisticc | ||
Ghosh et al. [26] | 32,574 incidents | Incident clearance time | Generalized F | ||
Chung et al. [53] | 3863 accidents | Duration time | Gamma and inverse Gaussian | ||
Semi-parametric hazard-based model | Hou et al. [27] | 2584 incidents | Clearance time | ||
Shi et al. [64] | 7203 incidents | Incident duration | |||
Regression and statistical tests | Log-linear models | Golob et al. [12] | 525 accidents | Incident duration | Log-normal |
Statistical tests | Giuliano [13] | 512 accidents | Response time + clearance time | Log-normal | |
Structural equation model | Lee et al. [11] | 3147 incidents | Incident clearance time | ||
OLS regression truncated regression | Zhang, Khattak [31] | 37,379 incidents | Event durationd | Log-normal or log-logistic distribution | |
Analysis of variance | Hojati et al. [36] | 4926 records | Incident duration time | Log-logistic and log-normale | |
Mechanism-based approach | Hou et al. [29] | 828 incidents | Response time | ||
Association rule learning algorithm | Lin et al. [65] | 999 accidents | Incident clearance time | ||
Binary probit and switching regression models | Ding et al. [51] | 1056 incidents | Response time and clearance time |
4 Traffic incident duration prediction
Method Category | Methodology | Data source | Duration time phase | Accuracy | |
---|---|---|---|---|---|
Regression model | Time sequential method (truncated regression model) | Khattak et al. [5] | 109 larger incidents | Duration time | Not test without available dataset |
Regression model | Garib et al. [6] | 205 incidents | Incident duration | 81% (adjusted R
2
) | |
Linear regression (LR) | Peeta et al. [7] | 835 crashes and 1176 debris | Clearance time | R2: 0.234 for crashes; 0.362 for debris | |
OLS regression models | Khattak et al. [32] | 59,804 incidents | Incident duration | Best MAPE: 37%a | |
A linear model with a stepwise regression | Yu, Xia [66] | 503 records | Incident duration | Acceptable (77.8% predictions have an error within 60 min) | |
Cluster-based log-normal distribution model | Weng et al. [67] | 2512 accidents | Accident duration | Best MAPE: 34.1% | |
Quantile Regression | Khattak et al. [68] | 85,000 incidents | Incident duration | RSME: 57.49 min | |
Fuzzy system | Fuzzy system model | Kim, Choi [69] | 2457 incidents | Incident service time | Average error: 0.3 min |
Fuzzy logic (FL) model | Wang et al. [70] | 457 records | Incident duration | Average performance | |
Fuzzy duration model | Dimitriou, Vlahogianni [71] | 1449 accidents | Accident duration | Best MAPE: 36%. | |
Classification Tree Method (CTM) | Decision tree | Ozbay, Kachroo [22] | 650 incidents | Clearance time | 60% less than 10 min |
Non-parametric regression and CTM | Smith, Smith [43] | 6828 accidents | Clearance time | Not good (correct rate 58%) | |
CTM | Knibbe et al. [72] | 1853 incidents | Incident duration time | Theoretical reliability: 65% | |
Hybrid tree-based quantile regression | He et al. [40] | 1245 incidents | Incident duration | MAPE: 49.1%. | |
M5P tree algorithm | Zhan et al. [15] | 2585 incidents | Lane clearance time | MAPE: 42.7%. | |
CTM | Chang, Chang [73] | 4697 cases | Incident duration | Accuracy of classification: 75.1%. | |
Artificial neural networks | FL and ANNs | Wang et al. [74] | 695 vehicle breakdowns | Incident duration | RMSE: about 20% |
ANNs | Wei, Lee [33] | 39 accidents | Accident duration | MAPE: 20%–30% | |
ANN-based models | Wei, Lee [16] | 24 incidents | Incident duration | MAPE mostly under 40%. | |
A sequential forecast based on two ANN-based models | Lee, Wei [17] | 39 accidents | Accident duration | The MAPE value at each time point is mostly under 29%. | |
Multiple LR; DT; ANN; SVM/RVM; K nearest neighbour (KNN) | Valenti et al. [19] | 237 incidents | Incident duration | MAPE of the five models: 34%–44%. | |
Four adaptive ANN-based models | Lopes et al. [56] | 10,762 incidents | Clearance time | Model 4: 72% incidents: <10 min error; 92%: <20 min error | |
Topic modelling and ANN-based models | Pereira et al. [45] | 10,139 accidents | Incident duration | A median error of 9.9 min in the best model | |
ANN models | Vlahogianni, Karlaftis [18] | 1449 accidents | Accident duration | Accuracy defined in the paper is about 10% | |
Bayesian ANNs | Park et al. [57] | 13,987 incidents | Incident duration | MAPE: 0.18–0.29. | |
Bayesian networks | Bayesian networks | Ozbay, Noyan [75] | 700 incidents | Incident clearance times | Accuracy of approximately 80% |
Probabilistic model based on a naïve Bayesian classifier (NBC) | Boyles et al. [8] | 2970 incidents | Incident duration | Classification is correct half of the time. | |
Bayesian decision model | Ji et al. [76] | 1853 incidents | Incident duration | Theoretical reliability of 74% | |
Tree-augmented NBC and a continuous model based on latent Gaussian NBC | Li, Cheng [77] | 2973 incidents | Incident duration | The frequency of the correct classification is below 0.5. | |
Bayesian network | Shen, Huang [78] | 2629 incidents | Incident duration | overall classification accuracy is 72.6% | |
hazard-based duration model | Time sequential procedure with HBDM | Qi, Teng [55] | 1660 incidents | Remaining incident duration | Accuracy increases with more information |
Log-logistic AFT model | Chung [58] | 4869 accidents | Accident duration | MAPE: 47%. | |
Log-logistic AFT model | Hu et al. [35] | 5362 incidents | Incident duration | MAPE: 43.7%. | |
Weibull AFT model | Kang, Fang [79] | 1327 incidents | Incident duration | MAPE: 43%. | |
KNN and Log-logistic AFT model | Araghi et al. [34] | 5362 incidents | Incident duration | MAPE: KNN: 41.1%; AFT: 43.7% | |
HBDM | Ji et al. [38] | 24,604 incidents | Clearance and arrival time | 39.68% of incident: <10 min error | |
Competing risk mixture HBDM | Li et al. [52] | 12,093 incidents | Incident duration | MAPE: 45% for >15 mins | |
G-component mixture model | Zou et al. [44] | 2584 incidents | Clearance time | MAPE: 39% | |
SVM | Ordered probit model and SVM | Zong et al. [80] | 3914 cases | Accident duration | MAPE: 22% |
SVM | Wu et al. [81] | 1853 incidents | Incident duration | Total accuracy: 70% | |
Combined/hybrid | Ordered probit model and a rule-based supplemental module | Lin et al. [10] | 22,495 incidents | Incident duration | Duration less than 60 min is 82.25% (within 10-min error) |
CTM and Rule-Based Tree Model (RBTM), DCM | Kim et al. [14] | 4 years’ worth of data | Incident duration | The overall confidence is more than 80%. | |
A hybrid model that consists of a RBTM, MultiNomial Logit model (MNL), and NBC | Kim, Chang [20] | 6765 records | Incident duration | Performed satisfactorily for incidents that last from 120 to 240 min | |
Combined M5P tree and HBDM | Lin et al. [54] | 602 accident records | Accident duration | MAPE: 36.2% for I-64 and 31.87% for I-190. |
4.1 Prediction methods
4.1.1 Single and combined models
4.1.2 Sequential and one-time models
4.2 Evaluation of prediction accuracy
5 Challenges and future work
Challenges | Potential methods | Previous research |
---|---|---|
Combining multiple data resources | Intelligent vehicle system (for example, eCall) | |
Traffic condition detection information | ||
Crowdsourcing technology | ||
Time sequential prediction model | Based on response term’s report | |
Based information from social media | Gu et al. [61] | |
Outlier prediction | Different models for different duration ranges | |
A time sequential prediction model | ||
Improvement of prediction methods | Machine Learning | |
Updated HBDM | Li et al. [46] et al. | |
Combining recovery times | Combine new data resource | Hojati et al. [23] |
Influence of unobserved factors | Randomness model |