1. A defined, the division of regimes B and C can be realized by further applying Isolation Forest to the flow-density points in regimes B and C to identify the major cluster of data points as regime B and consider the remaining scatter of points as regime C. Isolation Forest is adopted here for not only its computational efficiency but its capability of detecting novelties. Novelty detection here is defined as determining whether a new observation is an outlier given the training data whereas outlier detection only identifies outliers within the training set. Novelty detection is crucial in the context of this study because it enables the division of incoming data (i.e., data that do not belong to the training set) into regimes B and C and thus the resulting incident detection algorithm can be cross-validated and implemented for real-time application. It is worth noting that recurrent congestions may not be observed at some detector station locations;C H Hsiao;Journal of Transportation Engineering,1994
2. A wavelet-based freeway incident detection algorithm with adapting threshold parameters;Y S Jeong;Transportation Research Part C: Emerging Technologies,2011
3. Classification of freeway traffic patterns for incident detection using constructive probabilistic neural networks;X Jin;IEEE Transactions on Neural networks,2001
4. Development and adaptation of constructive probabilistic neural network in freeway incident detection;X Jin;Transportation Research Part C: Emerging Technologies,2002
5. Experimental properties of complexity in traffic flow;B S Kerner;Physical Review E,1996