Automatic traffic incident detection based on nFOIL
Jian Lu
1
, Shuyan Chen
⇑
, Wei Wang
2
, Bin Ran
School of Transportation, Southeast University, Nanjing 210096, China
article info
Keywords:
Automated incident detection
Inductive logic programming
nFOIL system
The area under ROC curve
Rare-class classification
Resampling
Ensemble learning
abstract
Traffic incidents inevitably cause traffic delay and deteriorate road safety conditions. Incidents are
increasing alongside the fast economic growth. Due to the rampant growth of traffic incidents, develop-
ing efficient and effective automated incident detection (AID) techniques has prompted a growing world-
wide interest. In this paper, the great efforts on developing a new approach to this problem based on
nFOIL, a novel inductive logic programming (ILP), are done. By way of illustration, a simulated traffic data
generated from Ayer Rajah Expressway (AYE) highway in Singapore and a real traffic data collected in
I-880 freeway in California are used to assess the detection performance of this approach, and perfor-
mance metrics includes detection rate, false alarm rate, mean time to detection, classification rate and
the area under Receiver Operating Characteristic (ROC) curve (AUC). For comparison, we conducted the
experiments on neural networks and support vector machine. The experimental results showed that
nFOIL is sensitive to the skewed distribution of positive and negative examples in the dataset, and we
make use of two different techniques, resampling and ensemble learning, to cope with highly skewed
data in the context of ILP classification problems and investigated the effect of them typicality on the
performance of AID model. It is concluded that ILP based AID approach are feasible, and have a favorable
performance compared to neural networks and support vector machines.
Ó 2011 Elsevier Ltd. All rights reserved.
1. Introduction
Traffic incidents are defined as unpredictable and non-recurring
events such as accidents, disabled vehicles, spilled loads, tempo-
rary maintenance and construction activities, signal and detector
malfunctions, and other special or unusual events that disrupt
the normal traffic flow and result in a capacity reduction of a facil-
ity (Srinivasan, Loo, & Cheu, 2003). With the rapid increase in
metropolitan population and other urbanization activities, a huge
demand has been imposed on metropolitan transportation system.
As a consequent, traffic operation conditions deteriorate and the
frequency of traffic incidents increases significantly. In the United
States, more than half of congestion on freeways is caused by inci-
dents and almost all congestion on rural freeways is involved with
incidents, road reconstruction, and maintenance (Tang & Gao,
2005).
The loss and influence caused by traffic incidents is proportional
to their durations. Early detection of incidents is vital for formulat-
ing effective response strategy, and the benefits to be derived from
early incident detection and quick response can drastically reduce
traffic delay, improve road safety and real-time traffic control
(Yuan & Cheu, 2003). The rampant growth in traffic incidents has
led to significant interest in the development of effective incident
detection techniques. Over the past decades, various automatic
incident detection (AID) techniques have been proposed to address
this problem, such as pattern recognition, time series analysis, Kal-
man filters, and forth on (Black & Sreedevi, 2006). Since 1990,
many issues concerning advanced AID algorithms emerged, includ-
ing neural network, fuzzy logic, support vector machine (SVM)
(Cheu, Srinivasan, & Teh, 2003; Yuan & Cheu, 2003), rough set
(Chen, Wang, & Qu, 2007), partial least squares regression (Chen,
Wang, et al., 2007; Wang et al., 2008), ensemble learning (Chen,
Wang, Qu, & Lu, 2007; Chen et al., 2009), and decision tree learning
(Chen & Wang, 2009), and these advanced AID algorithms are often
thought as the most promising methods.
However, all incident detection algorithms based on neural net-
works suffer from a number of shortcomings, including slow con-
vergence, heuristic determination of parameters for training
neural networks and possibility of getting stuck in a local minima
situation. Similarly, SVM also has some drawbacks, since the per-
formance of SVM depends on the choice of kernel function and sev-
eral parameters for SVM, while there is no simple approach for
these parameter tuning. Fuzzy logic incorporates human knowl-
edge by fuzzy rules and fuzzy membership functions, while the
determination of these rules and membership functions is one of
the key problems, which is often set artificially, thus the detection
performance is generally affected by subjective decisions.
0957-4174/$ - see front matter Ó 2011 Elsevier Ltd. All rights reserved.
doi:10.1016/j.eswa.2011.12.050
⇑
Corresponding author. Tel.: +86 25 83795356.
E-mail addresses: Lujian_1972@seu.edu.cn (J. Lu), chenshuyan@seu.edu.cn
(S. Chen), wangwei@seu.edu.cn (W. Wang), bran@wisc.edu (B. Ran).
1
Tel.: +86 25 83795645.
2
Tel.: +86 25 83794101.
Expert Systems with Applications 39 (2012) 6547–6556
Contents lists available at SciVerse ScienceDirect
Expert Systems with Applications
journal homepage: www.elsevier.com/locate/eswa