If the order size is over 5, the judgment is perfect. Lower-order edges have asymmetric ROC curves, unlike in human beings. Fixed-order edges do not guarantee regular ROC shapes despite the order size. However, when the edges are composed of various edge sizes, we can see curved DNA-PK inhibitor clinical trial symmetric ROC graphs, as shown in Figure 9(b). The order composition differs from 2 to 6, and the curves appear to be regular. High-order edges increase the hit rate, and low-order edges affect the false alarms. The point of interest

here is that the curves were regular even though the false positive cases were changed according to the range of random orders. The third edge configuration contains randomly combined edge orders. The overall performance is better than the second edge configuration (see Figure 9(c)). The hit rate converges early when the edge order range is larger. However, like the fixed order edges, the curves show asymmetric shapes. Through the judgment experiment, we validated that a random edge configuration is most adaptable to the human-like recognition memory model. Figure 9 ROC curves for familiarity judgment according to the edge configuration: (a) fixed-order edges, (b) random-order edges, and (c) random-order edges with random

combinations. 4.2.2. Pattern Completion Another functionality of familiarity in recognition memory is pattern completion. From the connectivity graph, we can see an inversely proportional relation between familiarity judgment and network connectivity. A high connectivity between edges hinders the memory from

judging new instances. Likewise, the property of network connectivity also influences the performance of pattern completion. We predicted that the number of activated edges and links enables a whole instance to be completed from partial input data. Among the eight attributes in the Reality Mining data, we randomly selected three attributes to assign missing values in the input data. We then tested whether the memory generates the missing values. Furthermore, we evaluated whether the generated values are identical to the original input data. The former result was assigned Carfilzomib as the completeness rate and the latter was the expectation. We drew the change in performance for both the completeness and expectation according to the edge configuration. Figure 10 shows the pattern completion performance. Similar to our assumption, the overall performance was aided by the network connectivity. In case of two fixed-order edges, the completeness and expectation rates were the highest. However, the performance decreased drastically as the order size increased. Random-order edges with a random combination also showed a similar trend. The network connectivity directly affected the performance. When the memory was composed of random-order edges, the pattern completion performance slowly changed according to the change in connectivity.

Incident duration, which can be defined as the

time difference between incident occurrence and incident site clearance [3–5], includes four time intervals or phases [6]: (1) incident detection/reporting time, (2) incident preparation/dispatching time, (3) travel time, and (4) clearance/treatment SAR131675 VEGFR Inhibitors time. This study investigates the influences of various traffic incident characteristics, such as temporal, road, incident-related, and environmental characteristics, on incident duration time using parametric hazard-based models and flexible parametric hazard-based duration models, to provide more suitable distribution for the base hazard function. The dataset used in this study was extracted from the Incident Reporting and Dispatching System in Beijing, and it contains the characteristics and duration times of incidents that occurred on the 3rd Ring expressway mainline in 2008. This paper begins with a literature review about previous research on incident duration analysis and prediction. This review is followed by details on flexible parametric hazard-based

model development. Next, the used data is described with the use of descriptive analyses of incident duration time and incident characteristics. The model estimation results and model parameter interpretation are then presented. This paper concludes with a summary of findings and directions for future research. 2. Literature Review Over the past few decades, many studies have been conducted to investigate appropriate approaches and techniques

for the estimation and prediction of traffic incident duration time, mainly on freeways. The most typical approaches include (1) regression methods [3, 7–9], (2) Bayesian classifier [10–12], (3) Decision trees and Classification trees [13, 14], (4) neural networks [15–17], (5) the discrete choice model [18], (6) the structure equation model [19], (7) probabilistic distribution analyses [20, 21], (8) support/relevance vector machines [22], and (9) hybrid methods [23]. These studies on traffic incident duration modeling have been summarized elsewhere [24, 25]. Several kinds of hazard-based models have been recently used to estimate the factors affecting traffic incident Brefeldin_A duration/clearance time or predict traffic incident duration/clearance time. The majority of studies on incident duration analysis have used parametric hazard-based models, that is, accelerated failure time (AFT) models, because of the following reasons: (1) the baseline hazard rate contributes to the understanding of the natural history of the incident through the manner in which the hazard rate changes over time; and (2) the AFT model allows for the estimation of an acceleration factor that can capture the direct effect of a specific factor on survival time [26].

4.1. UCI Data Set In our experiments, totally four UCI data sets are used, including 4-dimensional Iris, 13-dimensional Wine, 10-dimensional Glass, and 34-dimensional Tyrphostin AG-1478 clinical trial Ionosphere. There are 3 clusters in data set of Iris, each of which has 50 data patterns; 3 clusters in data set of Wine, which have 50, 60, and 68 data patterns; 6 clusters in data set of Glass, which have 30, 35, 40, 42, 36, and 31 separately; and 2 clusters in data set of Ionosphere, which have 226 and 125 data patterns. The validity indices of each

method are compared in Table 1. SP-FCM can identify compact groups compared to other algorithms when given the cluster number C. It can also be seen that SRCM and SP-FCM have more obvious advantages than FCM, RCM, and SCM. SP-FCM performs slightly better than SRCM in most cases due to the global search ability which enables it to converge to an optimum or near optimum solutions.

Moreover, shadowed set- and rough set-based clustering methods, namely, SP-FCM, SRCM, RCM, and SCM, perform better than FCM. It implies that the partition of approximation regions can reveal the nature of data structure and only the lower bound and boundary region of each cluster have positive contribution in the process of updating the prototypes. Table 1 Performance of FCM, RCM, SCM, SRCM, and SP-FCM on four UCI data sets. As usual, the number of clusters is implied by the nature of the problem. Here, with the shadowed sets involved, one can anticipate that the optimal number of clusters could be found. The fuzzification coefficient m can be optimized; however, it is common to assume a fixed value of 2.0, which associates with the form of the membership functions of

the generated clusters. For testing the SP-FCM algorithm, the rule C ≤ N1/2 is adopted, and the range of the expected cluster number can be set as (1) Iris, [Cmin = 2, Cmax = 12]; (2) Wine, [Cmin = 2, Cmax = 13]; (3) Glass, [Cmin = 2, Cmax = 14]; (4) Ionosphere, [Cmin = 2, Cmax = 16]. The swarm size is set as L = 20, the maximum iteration number of PSO T = 50, and, for cluster reduction, the cluster cardinality threshold ε = 10 and the attrition rate ρ = 0.1. In each cycle, we get the distribution of every cluster, remove Carfilzomib part of them according to their cardinality, and calculate the XB index, and the cluster number C varies from Cmax to Cmin . After ending the circulation, the partition with the lowest value is selected as the final result. Figure 2 presents the validity indices in the process of generating the optimal cluster number. Smaller values indicate more compact and well-separated clusters. The validity indices reach their minimum value at C = 3, 3, 6, and 2 separately, which correspond to the final cluster prototype and the best partition. Through the shadowed sets and PSO approaches, the influence of each boundary region to the formation of the prototypes and the clusters can be properly resolved.