Safety performance functions (SPFs) are essential analytical tools in the road safety field. The SPFs are statistical regression models used to predict crash counts by roadway facility type, crash type and severity. The national Highway Safety Manual (HSM) is a generic guidebook used for road safety evaluation and enhancement. In it, default SPFs, developed using negative binomial (NB) regression, are provided for multiple facility types and crash categories. Roadway agencies, whether public or private, may opt to not invest their resources in data collection and processing to develop own localized SPFs. Instead, the agencies may adopt the HSM's. However, the HSM's SPFs may not necessarily be applicable to any conditions. Hence, this research is focused on SPF transferability, specifically for rural divided multilane highway segments. Use of Bayesian informative priors to aid in the transferability of NB SPFs, developed for Florida, to California's conditions and vice versa is investigated. It is demonstrated that informative priors facilitate SPF transferability. Furthermore, NB SPFs are developed for Florida, Ohio, Illinois, Minnesota, California, Washington and North Carolina. That is to evaluate the transferability of each state's SPFs to the other states' conditions. The results indicate that Ohio, Illinois, Minnesota and California have SPFs that are transferable to conditions of each of the four states. Also, two methods are proposed for calibrating transferred SPFs to the destinations' conditions and are shown to outperform the SPF calibration methods in the road safety literature. Finally, a variety of modeling frameworks are proposed for developing and transferring SPFs of the seven aforementioned states to each state's data. Not a single model exhibits the best fit when transferred in all cases. However, the Tobit model, NB model and a hybrid model that coalesces the results of both perform the best in a substantial number of the transferred SPFs.

We examined the potential to improve the movement of freight using Truck Platooning Lane strategies on limited access highways in the State of Florida. In the First part of this research, we investigated the potential benefits from dedicating one lane from existing lanes for autonomous trucks only. In this regard, a general framework tool was developed to evaluate and compare different measurements (e.g., travel tim and emissions) to better assist decision makers to determine the most effective freight transportation strategy. Additionally, the travel time, level of service and emissions on Florida Strategic Intermodal System (SIS) were systematically analyzed using a VISSIM and MOVES simulation to determine if it can be improved. For the scenarios simulated in this investigation, the input included different patterns with a variety of peak hour volumes, truck percentages, speeds, and number of lanes. Additionally, the various total values of the resultant travel time, emissions and level of service for each SIS corridor were determined and calculated using a General Linear Model and then tabulated to reveal input patterns. The results showed that a truck platooning lane can significantly reduce the travel time and emissions of trucks. In the second part, we proposed using a The Analytic Hierarchy Process (AHP) method to evaluate the potential benefits of building a new lane for autonomous trucks. The AHP method was developed to include all possible measurements that can assist decision makers to select the best autonomous truck policy. The results of the AHP model showed that the safety criterion was significantly the most influential perspective per experts' opinions. The results showed that experts were more concerned about safety and environmental considerations than the initial cost associated with building a new lane.

As one of the most important Active Traffic Management strategies, Adaptive Traffic Signal Control (ATSC) helps improve traffic operation of signalized arterials and urban roads by adjusting the signal timing to accommodate real-time traffic conditions. Recently, with the rapid development of artificial intelligence, many researchers have employed deep reinforcement learning (DRL) algorithms to develop ATSCs. However, most of them are not practice-ready. The reasons are two-fold: first, they are not developed based on real-world traffic dynamics and most of them require the complete information of the entire traffic system. Second, their impact on traffic safety is always a concern by researchers and practitioners but remains unclear. Aiming at making the DRL-based ATSC more implementable, existing traffic detection systems on arterials were reviewed and investigated to provide high-quality data feeds to ATSCs. Specifically, a machine-learning frameworks were developed to improve the quality of and pedestrian and bicyclist's count data. Then, to evaluate the effectiveness of DRL-based ATSC on the real-world traffic dynamics, a decentralized network-level ATSC using multi-agent DRL was developed and evaluated in a simulated real-world network. The evaluation results confirmed that the proposed ATSC outperforms the actuated traffic signals in the field in terms of travel time reduction. To address the potential safety issue of DRL based ATSC, an ATSC algorithm optimizing simultaneously both traffic efficiency and safety was proposed based on multi-objective DRL. The developed ATSC was tested in a simulated real-world intersection and it successfully improved traffic safety without deteriorating efficiency. In conclusion, the proposed ATSCs are capable of effectively controlling real-world traffic and benefiting both traffic efficiency and safety.

Real-time crash risk prediction aims to predict the crash probabilities within a short time period, it is expected to play a crucial role in the advanced traffic management system. However, most of the existing studies only focused on freeways rather than urban arterials because of the complicated traffic environment of the arterials. This thesis proposes a long short-term memory convolutional neural network (LSTM-CNN) to predict the real-time crash risk at arterials. The advantage of this model is it can benefit from both LSTM and CNN. Specifically, LSTM captures the long-term dependency of the data while CNN extracts the time-invariant features. Four urban arterials in Orlando, FL are selected to conduct a case study. Different types of data are utilized to predict the crash risk, including traffic data, signal timing data, and weather data. Various data preparation techniques are applied also. In addition, the synthetic minority over-sampling technique (SMOTE) is used for oversampling the crash cases to address the data imbalance issue. The LSTM-CNN is fine-tuned on the training data and validated on the test data via different metrics. In the end, five other benchmarks models are also developed for model comparison, including Bayesian Logistics Regression, XGBoost, LSTM, CNN, and Sequential LSTM-CNN. Experimental results suggest that the proposed LSTM-CNN outperforms others in terms of Area Under the Curve (AUC) value, sensitivity, and false alarm rate. The findings of this thesis indicate the promising performance of using LSTM-CNN to predict real-time crash risk at arterials.

Providing accurate traffic speed prediction is essential for the success of Intelligent Transportation Systems (ITS) deployments. Accurate traffic speed prediction allows traffic managers take proper countermeasures when emergent changes happen in the transportation network. In this thesis, we present a computationally less expensive machine learning approach XGBoost to predict the future travel speed of a selected sub-network in Beijing's transportation network. We perform different experiments for predicting speed in the network from future 1 min to 20 min. We compare the XGBoost approach against other well-known machine learning and statistical models such as linear regression and decision tree, gradient boosting tree, and random forest regression models. Three metrics MAE, MAPE, and RMSE are used to evaluate the performance of the selected models. Our results show that XGBoost outperforms other models across different experiment conditions. Based on the prediction accuracy of different links, we find that the number of vehicles operating in a network also affect prediction performance. In addition, understanding individual mobility behavior is critical for modeling urban dynamics. It provides deeper insights on the generative mechanisms of human movements. Recently, different types of emerging data sources such as mobile phone call detail records, social media posts, GPS observations, and smart card transactions have been used to analyze individual mobility behavior. In this thesis, we report the spatio-temporal mobility behaviors using large-scale data collected from a ride-hailing service platform. Based on passenger-level travel data, we develop an algorithm to identify users' visited places and the functions of those places. To characterize temporal movement patterns, we reveal the differences in trip generation characteristics between commuting and non-commuting trips and the distribution of gap time between consecutive trips. To understand spatial mobility patterns, we observe the distribution of the number of visited place and their rank, the spatial distribution of residences and workplaces, and the distribution of travel distance and travel time. Our analysis highlights the differences in mobility patterns of the users of ride-hailing services, compared to the findings of existing mobility studies based on other data sources. Our study shows a tremendous potential of developing high-resolution individual-level mobility model that can predict the demand of emerging mobility services with high accuracy.

Pedestrians are regarded as Vulnerable Road Users (VRUs). Each year, thousands of pedestrians' deaths are caused by traffic crashes, which take up 16% of the total road fatalities and injuries in the U.S. (FHWA, 2018). Crashes can happen if there are interactions between VRUs and motorized transportation. And pedestrians' unexpected crossings, such as red-light violations at the signalized intersections, would expose them to motorized transportation and cause potential collisions. This thesis is intended to predict the pedestrians' red-light violation behaviors at the signalized crosswalks based on an LSTM (Long Short-term Memory) neural network. With video data collected from real traffic scenes, it is found that pedestrians that crossed during the red-light periods are more in danger of being struck by vehicles, from the perspective of Surrogate Safety Measures (SSMs). Pedestrians' features are generated using computer vision techniques. An LSTM model is used to predict pedestrians' red-light violations using these features. The experiment results at one signalized intersection show that the LSTM model achieves an accuracy of 91.6%. Drivers can be more prepared for these unexpected crossing pedestrians if the model is to be implemented in the vehicle-to-infrastructure (V2I) communication system.

Wrong way driving (WWD) is dangerous. Recent utilization of advanced WWD Intelligent Transportation Systems (ITS) countermeasures has demonstrated a reduction in WWD activities. Examples of these advanced WWD ITS countermeasures include Rectangular Flashing Beacons (RFBs) and Light Emitting Diodes (LEDs). Agencies need to decide which highway sites would be best to deploy such devices to be most cost effective while minimizing the WWD risk in the highway network. Previous UCF research developed a highway segment model for determining WWD hotspots on limited access facilities. This hotspot model was applied to toll road networks in the state of Florida. Also, UCF previous research developed an optimization algorithm which was integrated with the WWD hotspot model to provide a cost-effective deployment of WWD countermeasures for use by highway agencies. This thesis examines the transferability and scalability of the UCF WWD hotspot and optimization methodology to a Florida statewide network. Different Wrong Way Crash Risk (WWCR) hotspot models were tested, and the Poisson model was selected which uses four-exit segments and five years of WWD event data. Sixty-three segments with 169 exit ramps not currently equipped with ITS countermeasures were identified as hotspots. It was found that 96 of the 169 ramps chosen by the optimization were not identified in the hotspots, indicating an improved investment utilization of 56.8% compared to just using the hotspot model. Comparing the WWD detection and turnaround rankings of sites currently equipped with RFBs or LEDs to the optimization rankings indicated a significant monotonic association between optimization rankings and turnaround percentage and detection rankings, thereby verifying the accuracy of the optimization. By showing the transferability and scalability of the UCF WWD hotspot and optimization methodology, this thesis can assist transportation agencies in reducing WWD in a cost-effective manner saving lives and money.

This study attempts to explore all factors associated with bicycle motor-vehicle crashes at intersections in order to improve bicycle safety and bicycle activity. Factors such as exposure (bicycle and vehicle volumes), existing facilities (bike lanes, sidewalks, shared-use paths), geometric design (# of lanes, speed limit, medians, legs, roadway conditions), and land-use were collected and evaluated using Poisson, Zero-Inflated Poisson, and Negative Binomial models in SAS 9.4 software. Increasing the bicycle travel mode can have positive lasting effects on personal health, the environment, and improve traffic conditions. Deterrents that keep users from riding bicycles more are the lack of facilities and most importantly, safety concerns. Florida has consistently been a national leader in bicyclist deaths, which made this area a great candidate to study. Vehicle and bicycle volumes for 159 intersections in Orlando, Florida were collected and compared with crash data that was obtained. All existing facilities, geometric design properties, and land-uses for each intersection were collected for analysis. The results confirmed that an increase of motor-vehicles and bicyclists would increase the risk of a crash at an intersection. The presence of a keyhole lane (bike lane in-between a through and exclusive right turn lane), was shown to be statistically significant, and although it still had a positive correlation with injury risk, it had a much lower risk of crashes than a typical bike lane at intersections. The presence of a far shared path (more than 4 feet from the edge of curb) was shown to be statistically significant in decreasing the risk of crashes between bicycles and motor-vehicles at intersections. Institutional, agricultural, residential, government, and school land uses had positive correlations and were statistically significant with increasing activity of bicyclists at intersections. This study is unique because it uses actual bicycle volume as an exposure to determine the effects of bicycle safety and activity at intersections and not many others have done this. It is important for transportation planners and designers to use this information to design better complete streets in the future.

Traffic data are essential for public agencies to monitor the traffic condition of the roadway network in real-time. Recently, public agencies have implemented Bluetooth Detection Systems (BDS) on arterials to collect traffic data and purchased data directly from private sector vendors. However, the quality and reliability of the aforementioned two data sources are subject to rigorous evaluation. The thesis presents a study utilizing high-resolution GPS trajectories to evaluate data from HERE, one of the private sector data vendors, and BDS of arterial corridors in Orlando, Florida. The results showed that the accuracy and reliability of BDS data are better than private sector data, which might be credited to a better presentation of the bimodal traffic flow pattern on signalized arterials. In addition, another preliminary study aiming at improving the quality of private sector data was also demonstrated. Information about bimodal traffic flow extracted by a finite mixture model from historical BDS is employed to augment real-time private sector data by a Bayesian inference framework. The evaluation of the augmented data showed that the augmentation framework is effective for the most part of the studied corridor except for segments highly influenced by traffic from or to the expressway ramps.

This thesis presents different data mining/machine learning techniques to analyze the vulnerable road users' (i.e., pedestrian and bicycle) crashes by developing crash prediction models at macro-level. In this study, we developed data mining approach (i.e., decision tree regression (DTR) models) for both pedestrian and bicycle crash counts. To author knowledge, this is the first application of DTR models in the growing traffic safety literature at macro-level. The empirical analysis is based on the Statewide Traffic Analysis Zones (STAZ) level crash count data for both pedestrian and bicycle from the state of Florida for the year of 2010 to 2012. The model results highlight the most significant predictor variables for pedestrian and bicycle crash count in terms of three broad categories: traffic, roadway, and socio demographic characteristics. Furthermore, spatial predictor variables of neighboring STAZ were utilized along with the targeted STAZ variables in order to improve the prediction accuracy of both DTR models. The DTR model considering spatial predictor variables (spatial DTR model) were compared without considering spatial predictor variables (aspatial DTR model) and the models comparison results clearly found that spatial DTR model is superior model compared to aspatial DTR model in terms of prediction accuracy. Finally, this study contributed to the safety literature by applying three ensemble techniques (Bagging, Random Forest, and Boosting) in order to improve the prediction accuracy of weak learner (DTR models) for macro-level crash count. The model's estimation result revealed that all the ensemble technique performed better than the DTR model and the gradient boosting technique outperformed other competing ensemble technique in macro-level crash prediction model.