Sami Demiroluk

Adaptive Learning in Bayesian Networks for Incident Duration Prediction

The development of a practical model for incident management is investigated through Bayesian networks (BNs) in this study. BNs are capable of accurately predicting incident durations and can easily be incorporated into incident management activities of traffic management centers to improve the real-time decision-making process. Three structure learning algorithms were used to construct BN structures. They were estimated by using 2005 New Jersey incident data; the best-performing one was chosen for the incident duration prediction with the use of the 10-fold cross-validation method and the Bayesian information criterion statistic. To demonstrate the performance of Bayesian learning, the chosen model was fed by 2011 New Jersey incident data on a monthly and quarterly basis. Comparing the prediction results for 2011 data with and without adaptive learning showed that the developed BN had the capability to automatically adapt itself to future conditions by learning the patterns of new incidents and their respective durations.

Feature selection for ranking of most influential variables for evacuation behavior modeling across disasters

The extensive list of factors that affect the evacuee decision process makes it difficult to design effective surveys and to develop decision models with high predictive power. Regression models and significance levels can help identify relevant variables and overcome this problem to an extent. However, such approaches fall short of ranking these variables or recognizing the redundant ones. In this study, the use of a feature selection method was proposed to ensure that the selected features were relevant and not at the same time redundant. This method, called conditional mutual information maximization, consists of picking features at each step and minimizes the uncertainty in the decision conditional on the response of any feature already picked. As a case study, the variables influencing evacuation behavior in the Northern New Jersey Evacuation Survey were ranked and compared for disaster scenarios. To validate the method and to demonstrate how it compared with the traditional methods, logistic regression models were also estimated with the same data set. It was found that the top-ranked variables might be available through an existing database such as the U.S. census and some could be calculated on the basis of the threat type and government action. This fact can be useful for emergency planners when an evacuation survey for a study area is not readily available. Overall, the feature selection algorithm succeeds in identifying the most influential factors for all threat types. The suggested approach can help both preprocessing (e.g., defining a set of input variables) and postprocessing (e.g., identification of variables that should be kept) for behavioral modeling.

Modeling Salt Usage During Snowstorms

Snow can cause dangerous driving conditions by reducing pavement friction and covering road surface markings. Salt is widely used by highway maintenance managers in the United States to reduce the impact of snow or ice on traffic. For the development of long-term plans, especially for the next winter season, it is essential to know what factors affect salt usage and to determine the sufficient amount of salt needed in each depot location. This determination can be accomplished by estimating statistically robust models for salt usage prediction. In this study, historical data regarding storm characteristics and salt usage on the New Jersey Turnpike and Garden State Parkway were used to estimate those models. Linear models, hierarchical linear models, and hierarchical linear models with varying dispersion (HLVDs) were developed to predict the salt usage on the two highways. Results show that districts with higher average snow depth, longer storm duration, and lower average temperature were associated with greater salt usage. HLVD models were found to have the best predictive performance by including random parameters to account for unobserved spatial heterogeneity and by including fixed effects in the dispersion term. In addition, with the estimation of case-specific dispersion on the basis of storm characteristics, HLVD models could be used appropriately to estimate the upper bounds of salt usage, bounds that are not extremely large and could satisfy the salt demand in most cases. The findings of this study can provide highway authorities with valuable insights into the use of statistical models for more efficient inventory management of salt and other maintenance materials.

An efficient maintenance and spare parts inventory management software for ITS equipment

Intelligent Transportation Systems (ITS) has two major sub-components that depend on each other namely, vehicles and infrastructure. For the whole system to perform optimally and safely, both vehicle and infrastructure sub-components of ITS should be operating without major disruptions. In this paper, we focus on the better maintenance of ITS equipment to maximize overall system performance. Long-term down time of ITS equipment due to the unavailability of spare parts will not only increase personnel and repair time requirements, costs of replacement parts but also might lead to increased delays, poor air quality and fuel consumption. Therefore, timely availability of the spare parts of essential components of ITS equipment is essential within the inspection and maintenance procedures of ITS. The proposed spare parts inventory control model in this paper can determine the optimum levels of the safety stocks under probabilistic failure and availability assumptions for components of various ITS equipment. When this inventory control model is fully integrated into Rutgers Intelligent Transportation Systems Inspection and Maintenance Software (RITSIMS) [1], it will allow its users to efficiently manage DOTs ITS spare parts inventory using historic maintenance and inspection data that is being collected by respective databases of RITSIMS.