Doctor of Philosophy (PhD)
Civil and Environmental Engineering and Construction
First Committee Member
Second Committee Member
Third Committee Member
Fourth Committee Member
Fifth Committee Member
Number of Pages
Currently, there are both methodological and practical barriers that together preclude a substantial use of theoretically sound approaches, such as the ones recommended by the Highway Safety Manual (HSM), for traffic safety management. Although the state-of-the-art provides theoretically sound approaches such as the Empirical Bayes method, there are still various important capabilities missing. Methodological barriers include among others (i) lack of a theoretically sound approach for corridor-level network screening, (ii) lack of a comprehensive approach for estimation of Safety Performance Functions based on a simultaneous consideration of both crash patterns and associated explanatory variables, and (iii) lack of theoretically sound methods to forecast crash patterns at the regional level. In addition, the use of existing theoretically sound approaches such as the ones recommended by the HSM are associated with important practical barriers including 1) significant data integration requirements, 2) a special schema is needed to enable analysis using specialized software, 3) time-consuming and intensive processes are involved, 4) substantial technical knowledge is needed, 5) visualization capabilities are limited, and 6) coordination across various data owners is required.
Considering the above barriers, most practitioners use theoretically unsound methodologies to perform traffic safety analyses for highway safety improvement programs. This research proposes a single comprehensive framework to address all the above barriers to enable the use of theoretically sound methodologies for network wide traffic safety analyses. The proposed framework provides access through a single platform, Business Intelligence (BI), to theoretically sound methods and associated algorithms, data management and integration tools, and visualization capabilities. That is, the proposed BI framework provides methods and mechanisms to integrate and process data, generate advanced and theoretically sound analytics, and visualize results through intuitive and interactive web-based dashboards and maps.
The proposed BI framework integrates data using Extract-Load-Transform process and creates a traffic safety data warehouse. Algorithms are implemented to use the data warehouse for network screening analysis of roadway segments, intersections, ramps, and corridors. The methodology proposed and implemented here for corridor-level network screening represents an important expansion to the existing methods recommended by the HSM. Corridor-level network screening is important for decision makers because it enables to rank corridors rather than sites so as to provide homogenous infrastructure to minimize changes within relatively short distances. Improvements are recommended for long sections of roadways that could include multiple sites with the potential for safety improvements. Existing corridor screening methodologies use observed crash frequency as a performance measure which does not consider regression-to-the-mean bias. The proposed methodology uses expected crash frequency as a performance measure and searches corridors using a sliding window mechanism which addresses crash location reporting errors by considering the same section of roadway multiple times using overlapping windows.
The proposed BI framework includes a comprehensive methodology for the estimation of SPFs considering simultaneously local crash patterns and site characteristics. The current state-of-the-art uses predefined crash site types to create single clusters of data to generate regression models, SPFs, for the estimation of predicted crash frequency. It is highly unlikely for all crash sites within a single predefined cluster/type to have similar crash patterns and associated explanatory characteristics. That is, there could be sites within a cluster/type with different crash patterns and explanatory characteristics. Hence, assigning a single predefined SPF to all sites within a type is not necessarily the best approach to minimize the estimation error. To address this issue, a mathematical program was formulated to determine simultaneously cluster memberships for crash sites and the corresponding SPFs. Cluster memberships are determined using both crash patterns and associated explanatory variables. A solution algorithm coupling simulation annealing and maximum log likely estimation was implemented and tested. Results indicated that multiple SPFs for a crash and/or facility type can maximize the probability of observing the available data to increase accuracy and reliability. The estimated SPFs using the proposed approach were implemented within the BI framework for network screening. The results illustrate that the gain in predicted crashes provided by the SPFs translates into superior rankings for sites and corridors with the potential for safety improvements.
A performance-based safety program requires the forecasting, at the regional level, of safety performance measures and establish targets to reduce fatalities and serious injuries. This is in contrast to the analysis required for traffic safety management where forecasts are required at the site or corridor level. For regional level forecasting, historically, theoretically unsound methods such as extrapolation or simple moving-average models have been used. To address this issue, this study proposed deterministic and stochastic time series models to forecast performance measures for performance-based safety programs. Results indicated that stochastic time series, a seasonal autoregressive integrated moving average model, provides the required statistically sound forecasts.
In summary, the fundamental contributions of this research include: (i) a theoretically sound methodology for corridor level network screening, (ii) a comprehensive methodology for the estimation of local SPFs considering simultaneously crash patterns and associated explanatory variables, and (iii) a theoretically sound methodology to forecast performance measures to set realistic targets for performance-based safety programs. In addition, this study implemented and tested the above contributions along with existing algorithms for traffic safety network screening within a single BI platform. The result is a single web-based BI framework to enable integration and management of source data, generation of theoretically sound analyses, and visualization capabilities through intuitive dashboards, drilldown menus, and interactive maps.
Clusterwise Regression; Highway Safety Management; Network Screening; Performance Measures Forecasting; Safety Performance Function; Time Series
Civil Engineering | Transportation
Veeramisti, Naveen Kumar, "A Business Intelligence Framework for Network-level Traffic Safety Analyses" (2016). UNLV Theses, Dissertations, Professional Papers, and Capstones. 2911.