Recent research at the University of Central Florida involving crashes on Interstate-4 in Orlando, Florida has led to the creation of new statistical models capable of determining the crash risk on the freeway (Abdel-Aty et al., 2004; 2005, Pande and Abdel-Aty, 2006). These models are able to calculate the rear-end and lane-change crash risks along the freeway in real-time through the use of static information at various locations along the freeway as well as the real-time traffic data obtained by loop detectors. Since these models use real-time traffic data, they are capable of calculating rear-end and lane-change crash risk values as the traffic flow conditions are changing on the freeway. The objective of this study is to examine the potential benefits of variable speed limit implementation techniques for reducing the crash risk along the freeway. Variable speed limits is an ITS strategy that is typically used upstream of a queue in order to reduce the effects of congestion. By lowering the speeds of the vehicles approaching a queue, more time is given for the queue to dissipate from the front before it continues to grow from the back. This study uses variable speed limit strategies in a corridor-wide attempt to reduce rear-end and lane-change crash risks where speed differences between upstream and downstream vehicles are high. The idea of homogeneous speed zones was also introduced in this study to determine the distance over which variable speed limits should be implemented from a station of interest. This is unique since it is the first time a dynamic distance has been considered for variable speed limit implementation. Several VSL strategies were found to successfully reduce the rear-end and lane-change crash risks at low-volume traffic conditions (60% and 80% loading conditions). In every case, the most successful treatments involved the lowering of upstream speed limits by 5 mph and the raising of downstream speed limits by 5 mph. In the free-flow condition (60% loading), the best treatments involved the more liberal threshold for defining homogeneous speed zones (5 mph) and the more liberal implementation distance (entire speed zone), as well as a minimum time period of 10 minutes. This treatment was actually shown to significantly reduce the network travel time by 0.8%. It was also shown that this particular implementation strategy (lowering upstream, raising downstream) is wholly resistant to the effects of crash migration in the 60% loading scenario. In the condition approaching congestion (80% loading), the best treatment again involved the more liberal threshold for homogeneous speed zones (5 mph), yet the more conservative implementation distance (half the speed zone), along with a minimum time period of 5 minutes. This particular treatment arose as the best due to its unique capability to resist the increasing effects of crash migration in the 80% loading scenario. It was shown that the treatments implementing over half the speed zone were more robust against crash migration than other treatments. The best treatment exemplified the greatest benefit in reduced sections and the greatest resistance to crash migration in other sections. In the 80% loading scenario, the best treatment increased the network travel time by less than 0.4%, which is deemed acceptable. No treatment was found to successfully reduce the rear-end and lane-change crash risks in the congested traffic condition (90% loading). This is attributed to the fact that, in the congested state, the speed of vehicles is subject to the surrounding traffic conditions and not to the posted speed limit. Therefore, changing the posted speed limit does not affect the speed of vehicles in a desirable manner. These conclusions agree with Dilmore (2005).

Recent research has shown that incorporating roadway safety in transportation planning has been considered one of the active approaches to improve safety. Aggregate level analysis for predicting crash frequencies had been contemplated to be an important step in this process. As seen from the previous studies various categories of predictors at macro level (census blocks, traffic analysis zones, census tracts, wards, counties and states) have been exhausted to find appropriate correlation with crashes. This study contributes to this ongoing macro level road safety research by investigating various trip productions and attractions along with roadway characteristics within traffic analysis zones (TAZs) of four counties in the state of Florida. Crashes occurring in one thousand three hundred and forty-nine TAZs in Hillsborough, Citrus, Pasco, and Hernando counties during the years 2005 and 2006 were examined in this study. Selected counties were representative from both urban and rural environments. To understand the prevalence of various trip attraction and production rates per TAZ the Euclidian distances between the centroid of a TAZ containing a particular crash and the centroid of the ZIP area containing the at fault driver's home address for that particular crash was calculated. It was found that almost all crashes in Hernando and Citrus County for the years 2005-2006 took place in about 27 miles radius centering at the at-fault drivers' home. Also about sixty-two percent of crashes occurred approximately at a distance of between 2 and 10 miles from the homes of drivers who were at fault in those crashes. These results gave an indication that home based trips may be more associated with crashes and later trip related model estimates which were found significant at 95% confidence level complied with this hypothesized idea. Previous aggregate level road safety studies widely addressed negative binomial distribution of crashes. Properties like non-negative integer counts, non-normal distribution, over-dispersion in the data have increased suitability of applying the negative binomial technique and has been selected to build crash prediction models in this research. Four response variables which were aggregated at TAZ-level were total number of crashes, severe (fatal and severe injury) crashes, total crashes during peak hours, and pedestrian and bicycle related crashes. For each response separate models were estimated using four different sets of predictors which are i) various trip variables, ii) total trip production and total trip attraction, iii) road characteristics, and iv) finally considering all predictors into the model. It was found that the total crash model and peak hour crash model were best estimated by the total trip productions and total trip attractions. On the basis of log-likelihoods, deviance value/degree of freedom, and Pearson Chi-square value/degree of freedom, the severe crash model was best fit by the trip related variables only and pedestrian and bicycle related crash model was best fit by the road related variables only. The significant trip related variables in the severe crash models were home-based work attractions, home-based shop attractions, light truck productions, heavy truck productions, and external-internal attractions. Only two variables- sum of roadway segment lengths with 35 mph speed limit and number of intersections per TAZ were found significant for pedestrian and bicycle related crash model developed using road characteristics only. The 1349 TAZs were grouped into three different clusters based on the quartile distribution of the trip generations and were termed as less-tripped, moderately-tripped, and highly-tripped TAZs. It was hypothesized that separate models developed for these clusters would provide a better fit as the clustering process increases the homogeneity within a cluster. The cluster models were re-run using the significant predictors attained from the joint models and were compared with the previous sets of models. However, the differences in the model fits (in terms of Alkaike's Information Criterion values) were not significant. This study points to different approaches when predicting crashes at the zonal level. This research is thought to add to the literature on macro level crash modeling research by considering various trip related data into account as previous studies in zone level safety have not explicitly considered trip data as explanatory covariates.

This thesis focuses on evaluating the appropriate speed at which vehicles should travel under different traffic conditions on freeways and its impact on crash frequency. The common belief is that the lower speed results in fewer crashes as reduced speed provides drivers with more time to react effectively and avoid collisions. However, this perspective overlooks the interplay among traffic speed, average spacing between consecutive vehicles, and the distance available for stopping a vehicle. Hence, we propose a safety parameter termed ‘Safety Correlate' (SCORE), which is defined as the proportion of average spacing relative to the stopping distance. To determine the relationship between SCORE and crash frequency, data from 366 4-lane urban freeway segments located in Virginia was analyzed and a Random-effects Poisson Lognormal model was developed. The obtained result indicated that the safety parameter SCORE is negatively associated with the annual hourly crash frequency, implying that the lesser the average spacing as a proportion of the stopping distance while traffic flow remains constant, the more frequent will be the crashes. Additionally, this research presents an application of SCORE in setting variable speed limits under various traffic flows. Overall, the study results provide valuable insights by investigating SCORE to improve traffic safety. Also, this research would help practitioners and policymakers to incorporate safety aspects while setting variable speed limits on freeways.

This research evaluates Intelligent Transportation System (ITS) implementation strategies to improve the safety of a freeway once a potential of a crash is detected. Among these strategies are Variable Speed Limit (VSL) and ramp metering. VSL are ITS devices that are commonly used to calm traffic in an attempt to relieve congestion and enhance throughput. With proper use, VSL can be more cost effective than adding more lanes. In addition to maximizing the capacity of a roadway, a different aspect of VSL can be realized by the potential of improving traffic safety. Through the use of multiple microscopic traffic simulations, best practices can be determined, and a final recommendation can be made. Ramp metering is a method to control the amount of traffic flow entering from on-ramps to achieve a better efficiency of the freeway. It can also have a potential benefit in improving the safety of the freeway. This thesis pursues the goal of a best-case implementation of VSL. Two loading scenarios, a fully loaded case (90% of ramp maximums) and an off-peak loading case (60% of ramp maximums), at multiple stations with multiple implementation methods are strategically attempted until a best-case implementation is found. The final recommendation for the off-peak loading is a 15 mph speed reduction for 2 miles upstream and a 15 mph increase in speed for the 2 miles downstream of the detector that shows a high crash potential. The speed change is to be implemented in 5 mph increments every 10 minutes. The recommended case is found to reduce relative crash potential from .065 to -.292, as measured by a high-speed crash prediction algorithm (Abdel-Aty et al. 2005). A possibility of crash migration to downstream and upstream locations was observed, however, the safety and efficiency benefits far outweigh the crash migration potential. No final recommendation is made for the use of VSL in the fully loaded case (low-speed case); however, ramp metering indicated a promising potential for safety improvement.

The present study focused on evaluating some crash prediction models for two lane highways on Brazilian conditions. Also, the transferability of models was considered, specifically by means of a comparison between Brazil, HSM and Florida. The analysis of two lane highways crash prediction models was promising when the road characteristics were well known and there was not much difference from base conditions. This conclusion was attained regarding the comparison of results for all segments, non-curved segments and curved segments, confirming that a transferred model can be used with caution. In addition, two novel models for Brazilian two-lane highways segments were estimated. The model developed showed better results for non-curved segments in the calibration/validation sample. Thus, for a general analysis purpose of non-curved segments this model is recommended. Finally, there are many factors that could not be measured by these models and reflects road safety various condition. Even so, the study of crash predict models in Brazilian context could provide a better start point in safety road analysis. / O foco desta pesquisa foi avaliar a aplicação de alguns modelos de previsão de acidentes em rodovias de pista simples de três estados brasileiros. Ainda, a transferabilidade destes modelos foi abordada, especificamente por meio de uma comparação entre características do Brasil, Florida e aquelas recomendadas pelo Highway Safety Manual. O uso dos distintos modelos se mostrou promissor para situações nas quais as características da via se mantiveram semelhantes às condições para as quais os modelos foram desenvolvidos. A avaliação foi empreendida para todos os segmentos homogêneos, separados posteriormente segundo a existência de curvas horizontais. Adicionalmente, dois novos modelos foram equacionados para a amostra brasileira. O modelo de previsão de acidentes desenvolvido apresentou melhores medidas de desempenho para segmentos sem curvas horizontais, sendo recomendável para previsão de acidentes em análises preliminares. Por fim, foi constatado que outros fatores não contemplados pelos modelos podem ter impactado as condições de segurança dos locais estudados. Ainda assim, essa pesquisa representa no contexto do Brasil um ponto de partida em análises relacionadas à segurança de rodovias de pista simples.

The objective of this research is to evaluate and introduce a new methodology regarding rural highway safety. Current practices rely on crash prediction models that utilize specific explanatory variables, whereas the depository of knowledge for past research is the Highway Safety Manual (HSM). Most of the prediction models in the HSM identify the effect of individual geometric elements on crash occurrence and consider their combination in a multiplicative manner, where each effect is multiplied with others to determine their combined influence. The concepts of 3-dimesnional (3-D) representation of the roadway surface have also been explored in the past aiming to model the highway structure and optimize the roadway alignment. The use of differential geometry on utilizing the 3-D roadway surface in order to understand how new metrics can be used to identify and express roadway geometric elements has been recently utilized and indicated that this may be a new approach in representing the combined effects of all geometry features into single variables. This research will further explore this potential and examine the possibility to utilize 3-D differential geometry in representing the roadway surface and utilize its associated metrics to consider the combined effect of roadway features on crashes. It is anticipated that a series of single metrics could be used that would combine horizontal and vertical alignment features and eventually predict roadway crashes in a more robust manner. It should be also noted that that the main purpose of this research is not to simply suggest predictive crash models, but to prove in a statistically concrete manner that 3-D metrics of differential geometry, e.g. Gaussian Curvature and Mean Curvature can assist in analyzing highway design and safety. Therefore, the value of this research is oriented towards the proof of concept of the link between 3-D geometry in highway design and safety. This thesis presents the steps and rationale of the procedure that is followed in order to complete the proposed research. Finally, the results of the suggested methodology are compared with the ones that would be derived from the, state-of-the-art, Interactive Highway Safety Design Model (IHSDM), which is essentially the software that is currently used and based on the findings of the HSM.

The present study focused on evaluating some crash prediction models for two lane highways on Brazilian conditions. Also, the transferability of models was considered, specifically by means of a comparison between Brazil, HSM and Florida. The analysis of two lane highways crash prediction models was promising when the road characteristics were well known and there was not much difference from base conditions. This conclusion was attained regarding the comparison of results for all segments, non-curved segments and curved segments, confirming that a transferred model can be used with caution. In addition, two novel models for Brazilian two-lane highways segments were estimated. The model developed showed better results for non-curved segments in the calibration/validation sample. Thus, for a general analysis purpose of non-curved segments this model is recommended. Finally, there are many factors that could not be measured by these models and reflects road safety various condition. Even so, the study of crash predict models in Brazilian context could provide a better start point in safety road analysis. / O foco desta pesquisa foi avaliar a aplicação de alguns modelos de previsão de acidentes em rodovias de pista simples de três estados brasileiros. Ainda, a transferabilidade destes modelos foi abordada, especificamente por meio de uma comparação entre características do Brasil, Florida e aquelas recomendadas pelo Highway Safety Manual. O uso dos distintos modelos se mostrou promissor para situações nas quais as características da via se mantiveram semelhantes às condições para as quais os modelos foram desenvolvidos. A avaliação foi empreendida para todos os segmentos homogêneos, separados posteriormente segundo a existência de curvas horizontais. Adicionalmente, dois novos modelos foram equacionados para a amostra brasileira. O modelo de previsão de acidentes desenvolvido apresentou melhores medidas de desempenho para segmentos sem curvas horizontais, sendo recomendável para previsão de acidentes em análises preliminares. Por fim, foi constatado que outros fatores não contemplados pelos modelos podem ter impactado as condições de segurança dos locais estudados. Ainda assim, essa pesquisa representa no contexto do Brasil um ponto de partida em análises relacionadas à segurança de rodovias de pista simples.

Indiana University-Purdue University Indianapolis (IUPUI) / Automotive technology has grown from strength to strength in the recent years. The main focus of research in the near past and the immediate future are autonomous vehicles. Autonomous vehicles range from level 1 to level 5, depending on the percentage of machine intervention while driving. To make a smooth transition from human driving and machine intervention, the prediction of human driving behavior is critical. This thesis is a subset of driving behavior prediction. The objective of this thesis is to predict the possibility of crash and implement an appropriate active safety system to prevent the same. The prediction of crash requires data of transition between lanes, and speed ranges. This is achieved through a variation of hidden Markov model. With the crash prediction and analysis of the Markov models, the required ADAS system is activated. The above concept is divided into sections and an algorithm was developed. The algorithm is then scripted into MATLAB for simulation. The results of the simulation is recorded and analyzed to prove the idea.

Real-time crash prediction is a complex task, since there is no existing framework to predict crash likelihood, types, and severity together along with a real-time traffic management strategy. Developing such a framework presents various challenges, including not independent and identically distributed data, imbalanced data, large model size, high computational cost, missing data, sensitivity vs. false alarm rate (FAR) trade-offs, estimation of traffic restoration time after crash occurrence, and real-world deployment strategy. A novel spatial ensemble distillation learning modeling technique is proposed to address these challenges. First, large-scale real-time data were used to develop a crash likelihood prediction model. Second, the proposed crash likelihood model's viability in predicting specific crash types was tested for real-world applications. Third, the framework was extended to predict crash severity in real-time, categorizing crashes into four levels. The results demonstrated strong performance with sensitivities of 90.35%, 94.80%, and 84.23% for all crashes, rear-end crashes, and sideswipe/angle crashes, and 83.32%, 81.25%, 83.08%, and 84.59% for fatal, severe, minor injury, and PDO crashes, respectively, all while remaining very low FARs. This methodology can also reduce model size, lower computation costs, improve sensitivity, and decrease FAR. These results will be used by traffic management center for taking measures to prevent crashes in real-time through active traffic management strategies. The framework was further extended for efficient traffic management after any crash occurrence despite adopting these strategies. Particularly, the framework was extended to predict the traffic state after a crash, predict the traffic restoration time based on the estimated post-crash traffic state, and apply a three-step validation technique to evaluate the performance of the developed approach. Finally, real-world deployment strategies of the proposed methodologies for real-time crash prediction along with their types and severities and real-time post-crash management are discussed. Overall, the methodologies presented in this dissertation offer multifaceted novel contributions and have excellent potential to reduce fatalities and injuries.

The unavailability of crash-related data has been a long lasting challenge in Brazil. In addition to the poor implementation and follow-up of road safety strategies, this drawback has hampered the development of studies that could contribute to national goals toward road safety. In contrast, developed countries have built their effective strategies on solid data basis, therefore, investing a considerable time and money in obtaining and creating pertinent information. In this research, we aim to assess the potential impacts of supplementary data on spatial model performance and the suitability of different spatial modeling approaches on crash prediction. The intention is to notify the authorities in Brazil and other developing countries, about the importance of having appropriate data. In this thesis, we set two specific objectives: (I) to investigate the spatial model prediction accuracy at unsampled subzones; (II) to evaluate the performance of spatial data analysis approaches on crash prediction. Firstly, we carry out a benchmarking based on Geographically Weighted Regression (GWR) models developed for Flanders, Belgium, and São Paulo, Brazil. Models are developed for two modes of transport: active (i.e. pedestrians and cyclists) and motorized transport (i.e. motorized vehicles occupants). Subsequently, we apply the repeated holdout method on the Flemish models, introducing two GWR validation approaches, named GWR holdout1 and GWR holdout2. While the former is based on the local coefficient estimates derived from the neighboring subzones and measures of the explanatory variables for the validation subzones, the latter uses the casualty estimates of the neighboring subzones directly to estimate outcomes for the missing subzones. Lastly, we compare the performance of GWR models with Mean Imputation (MEI), K-Nearest Neighbor (KNN) and Kriging with External Drift (KED). Findings showed that by adding the supplementary data, reductions of 20% and 25% for motorized transport, and 25% and 35% for active transport resulted in corrected Akaike Information Criterion (AICc) and Mean Squared Prediction Errors (MSPE), respectively. From a practical perspective, the results could help us identify hotspots and prioritize data collection strategies besides identify, implement and enforce appropriate countermeasures. Concerning the spatial approaches, GWR holdout2 out performed all other techniques and proved that GWR is an appropriate spatial technique for both prediction and impact analyses. Especially in countries where data availability has been an issue, this validation framework allows casualties or crash frequencies to be estimated while effectively capturing the spatial variation of the data. / A indisponibilidade de variáveis explicativas de acidentes de trânsito tem sido um desafio duradouro no Brasil. Além da má implementação e acompanhamento de estratégias de segurança viária, esse inconveniente tem dificultado o desenvolvimento de estudos que poderiam contribuir com as metas nacionais de segurança no trânsito. Em contraste, países desenvolvidos tem construído suas estratégias efetivas com base em dados sólidos, e portanto, investindo tempo e dinheiro consideráveis na obtenção e criação de informações pertinentes. O objetivo dessa pesquisa é avaliar os possíveis impactos de dados suplementares sobre o desempenho de modelos espaciais, e a adequação de diferentes abordagens de modelagem espacial na previsão de acidentes. A intenção é notificar as autoridades brasileiras e de outros países em desenvolvimento sobre a importância de dados adequados. Nesta tese, foram definidos dois objetivos específicos: (I) investigar a acurácia do modelo espacial em subzonas sem amostragem; (II) avaliar o desempenho de técnicas de análise espacial de dados na previsão de acidentes. Primeiramente, foi realizado um estudo comparativo, baseado em modelos desenvolvidos para Flandres (Bélgica) e São Paulo (Brasil), através do método de Regressão Geograficamente Ponderada (RGP). Os modelos foram desenvolvidos para dois modos de transporte: ativos (pedestres e ciclistas) e motorizados (ocupantes de veículos motorizados). Subsequentemente, foi aplicado o método de holdout repetido nos modelos Flamengos, introduzindo duas abordagens de validação para GWR, denominados RGP holdout1 e RGP holdout2. Enquanto o primeiro é baseado nas estimativas de coeficientes locais derivados das subzonas vizinhas e medidas das variáveis explicativas para as subzonas de validação, o último usa as estimativas de acidentes das subzonas vizinhas, diretamente, para estimar os resultados para as subzonas ausentes. Por fim, foi comparado o desempenho de modelos RGP e outras abordagens, tais como Imputação pela Média de dados faltantes (IM), K-vizinhos mais próximos (KNN) e Krigagem com Deriva Externa (KDE). Os resultados mostraram que, adicionando os dados suplementares, reduções de 20% e 25% para o transporte motorizado, e 25% e 35% para o transporte ativo, foram resultantes em termos de Critério de Informação de Akaike corrigido (AICc) e Erro Quadrático Médio da Predição (EQMP), respectivamente. Do ponto de vista prático, os resultados poderiam ajudar a identificar hotspots e priorizar estratégias de coleta de dados, além de identificar, implementar e aplicar contramedidas adequadas. No que diz respeito às abordagens espaciais, RGP holdout2 teve melhor desempenho em relação a todas as outras técnicas e, provou que a RGP é uma técnica espacial apropriada para ambas as análises de previsão e impactos. Especialmente em países onde a disponibilidade de dados tem sido um problema, essa estrutura de validação permite que as acidentes sejam estimados enquanto, capturando efetivamente a variação espacial dos dados.