Evaluation of Intelligent Transportation System Operations Using Logistic Regression Models

Transcription

1 Evaluation of Intelligent Transportation System Operations Using Logistic Regression Models Real-time incident management Introduction Intelligent Transportation System (ITS) operations data and ITS operations has become one of the core business functions for U.S. transportation operations data can be used agencies. ITS provides real-time situational awareness and disseminates alerts to to develop ITS Operations motorists on a variety of concerns, including traffic incidents, special events, work Evaluations Models (ITSOEMs) zone traffic control, expected delays, estimated travel times, diversion routes, and that are capable of quantifying lane closures. Recent U.S. Department of Transportation (USDOT) findings indicate that the use of ITS systems such as the operations value of dynamic dynamic message signs (DMS), integrated message signs and their safety service patrols (SSP), closed-circuit television (CCTV) surveillance systems, complementing ITS surveillance and vehicle detection systems (VDS) during recurring and nonrecurring congestion events considerably enhance safety and detection systems, such as and result in optimal use of the available closed-circuit television, safety capacity on transportation corridors. 1 Though ITS operations continue to gain service patrol, and computeraided dispatch systems. The ITS role of ITS in transportation system op- prominence, very few studies attempt to capture and quantitatively evaluate the erations in general and their effectiveness during incidents and other recurring Operations Evaluation Models and nonrecurring congestion events in developed in this study provide particular. Further, because of competing and declining budget challenges, meaningful and conclusive transportation system operators are often challenged to justify ITS investments in a inferences. quantifiable manner. A literature review 2,3 confirms that a few studies that attempt to evaluate ITS impacts during incidents and nonrecurring BY GUMMADA MURTHY, P.E., JIAN JOHN LU, PH.D., P.E., congestion events AND LAKSHMINARAYAN RAJARAM, PH.D. have reported lack of availability of granular data on ITS operations and incident management and clearly demonstrated the need for analytical models to perform quantifiable assessment of ITS operations. Such analytical models will enable transportation system operators to make a comparative performance assessment of existing ITS operations, conduct ITS gap analysis, and develop performance-based investment plans leading to effective ITS deployment decisions. Approach The Virginia I-95 corridor was selected as the study corridor (Figure 1), and the required operations data were retrieved from the ITS and incident management operations programs administered by the Virginia Department of Transportation (VDOT). VDOT planning methods 4 and criteria were used, and the I-95 study corridor was delineated into urban and transition segments. For each segment of the I-95 corridor, the real-time operations data were extracted for critical analysis time periods. Discussion with VDOT officials confirmed that DMS and other ITS operations are prominently used in the urban and transition segments of the I-95 study corridor. Through additional discussions with VDOT operations engineers, the analysis scenarios have been selected to better reflect and capture the critical corridor operations characteristics for the selected analysis periods, such as HOV lane operations, freight and passenger mobility trends, and prevalent maintenance and work zone situational conditions. The DMS and ITS operation data elements in this study include DMS messages, DMS location, direction, message start time and end time, message duration, and message type. ITS detection systems data include CCTV, SSP, and Virginia State Police Computer Aided Dispatch (VSP-CAD) system detections. The incident management data include incident type, lanes impacted, incident start time and end time, incident duration, incident location, direction, involvement of trucks in incidents, hazmat spills, weather im- 40 ITE JOURNAL / MARCH 2013

2 pacts, and incidents in work zones. Using these data, the analysis variables required for this study are identified and a set of quantitative ITS Operations Evaluation Models are developed for the I-95 study corridor. The first model () is designed to determine the probability that a DMS message occurs given the descriptors of the detections and incidents. The second,, is meant to determine the measurable influence of disseminated messages in the deterrence of secondary incidents on the I-95 study corridor. The models developed are expected to establish meaningful relationships between the analysis variables and provide measurable evaluation of DMS and ITS detection system operations during incident management. Figure 1. I-95 study corridor. (Note: MP = mile post) Use of Logistic Regression to Develop and Logistic regression is a statistical modeling method that describes the relationship between the categorical response variable and one or more continuous and/ or categorical explanatory variables. Several studies 5 9 confirmed that the logistic regression model is useful in examining and developing analytical models based on the relationship between the categorical response variable and the categorical and/or continuous explanatory variables. Review of the VDOT ITS operations and incident management data indicated that the DMS operations data, ITS detection data from CCTV, VSP-CAD, and SSP activations data can be coded as categorical variables based on the presence or absence of the DMS and detection system activations during the occurrence of incidents on the I-95 study corridor. Similarly, data on incidents that occur on the I-95 study corridor can also be coded as categorical variables. The other variables considered in this study for the development of IT- SOEMs, such as incident duration (in minutes) and distance of DMS location to the scene of incident (in miles) are clearly continuous data. Based on this, the data analyzed for this study are converted to categorical format and used to develop logistic regression and models. The Logistic Regression model, which is based on the cumulative logistic probability function F(Z i ), is specified as follows by Munley, 8 Hosmer et al., 9 and Sinha et al. 10 : Z i = log ( 1 ) = (β 0 + β 1 X 1 + β 2 X 2 + β 3 X 3 + β 4 X 4 + β 5 X 5 (1) + β 6 X 6..+ β p X p ) In the above equation the β 1, β 2, etc. are the model coefficients representing the Urban MP 155-MP170 Transition MP 125-MP155 numeric weight or influence of the independent variable on the dependent variable. Further, the exponent of the coefficients β 1, β 2, is called the odds ratio and is critical for the study analysis. The odds ratio approximates how much more likely it is for the outcome to be present given the presence of the independent variable in the logistic regression. Therefore, the logistic model transforms the focus of predicting ITE JOURNAL / MARCH

3 Table 1. I-95 study corridor ITS operations data and incident data attributes. DMS Attributes DMS Message Date DMS Sign ID DMS Location Mile Marker Direction (NB/SB) Device Location Message Duration Message Types: Accident, Delay, Roadwork, Major Event, Road Closed ITSOEM Model Variables D CCTV D VSPCAD D SSP VDOT Data Source VA Traffic and ATMS VA Traffic and ATMS VA Traffic and ATMS Incident Date Incident Attributes Incident Location Mile Marker Time of Day Incident Duration Incident ID Direction (NB/SB) Incident Detection: SSP, VSP, CCTV, Others Table 2. ITSOEM model variables. Incident Type: Trucks, Hazmat, Fatalities, and Roadway Infrastructure Damage ITSOEM Model Variables Description Incident detections resulting from 24/7 closed circuit television camera surveillance detection operations This is a categorical variable and takes 0/1 value. Incident detections resulting from 24/7 Virginia State Police Computer Aided Dispatch (VSPCAD) operations. This is a categorical variable and takes 0/1 value. Incident detections resulting from continuous roving safety service patrol (SSP) operations. This is a categorical variable and takes 0/1 value. I HIGH VA Traffic High-profile complex incidents that cause severe delays and road closures. This is a categorical variable and takes 0/1 value. I MAJOR VA Traffic Major-profile incidents severe enough to block one or more lanes for a considerable duration. This is a categorical variable and takes 0/1 value. I MINOR VA Traffic Minor-profile incidents resulting from fender-bendertype crashes, disabled vehicles, and debris spills. This is a categorical variable and takes 0/1 value. VA Traffic Incident start time to close time. This is a continuous variable. D DMS VA Traffic Distance of dynamic message sign (DMS) location to the incident location (miles). This is a continuous variable. DMS APP VOL VA Traffic Hourly approach volume to DMS location (vehicles per hour). This is a continuous variable. DMS MESSAGES ATMS Messages posted on DMS. This is the dependent variable for. This is a categorical dependent variable and takes 0/1 value. I SECONDARY VA Traffic It is expected that as incident duration increases there is a strong likelihood of secondary crash occurrence. This variable is the dependent or response variable for Models. This is a categorical dependent variable and takes 0/1 value. probabilities within a (0/1) range to the objective of predicting the odds of an event occurring within the range. 9 Logistic regression has become one of the most prominent and accepted method of analysis of binary outcome variables, attributed to the ease of interpretation of results from logistic regression fitted models and the power of estimated odds ratios in explaining the influence of independent variables on the dependent variables Data The 2009 data for the urban and transition segments of the study corridor derived from VDOT s databases consist of granular operations details for 3,425 incident records. These VDOT data also include pertinent DMS and ITS operations data for each record. VDOT s DMS and ITS operations data and incident management data are further synthesized using date, time, and space attributes as shown in Table 1. Considering the VDOT operations manual, 13 the following incident operations criteria for the I-95 study corridor were used to denote secondary incidents: (1) all incident records in the final data sets prepared for the ITSOEM development are initially considered as primary incidents, (2) the demarcated secondary incident occurs within 4.5 miles from the primary incident, (3) the demarcated secondary incident occurs within 30 minutes from the end time of the primary incident, (4) the secondary incident is in the same cardinal direction as the primary incident, and (5) the secondary incident is at an upstream location to the primary incident. Development of and Models Two principal underpinnings form the logistic basis for developing the and models. For developing, the dependent variable DMS Messages is in a binary (0/1) form, and takes the value of 0 for incident detections not followed by message dissemination on DMS, and 1 for detections followed by message dissemination on DMS. The IT- SOEM1 models sole purpose is designed to determine the probability that a DMS message occurs, given the descriptors of the detections and incidents. Similarly, for IT- 42 ITE JOURNAL / MARCH 2013

4 SOEM2 the dependent variable secondary incidents also takes binary (0/1) form, with the value 0 for incidents not followed by secondary incidents and the value of 1 for primary incidents followed by secondary incidents. The dependent and independent variables considered for inclusion into the and logistic regression models are described in Table 2. The analysis scenarios considered for ITSOEM model building are established in such a manner that the traffic, freight, and high-occupancy lane operations that define the traffic and regional mobility attributes of the I-95 study corridor are captured by the ITSOEMs. In this paper, the results from and models for 5 a.m. to 1 p.m. analysis scenarios are used to show how these models are applied to quantify ITS operations during incident management on the I-95 study corridor. ITSOEM binary logistic regression models are developed using SPSS 19 statistical analysis software and by adopting block-wise logistic regression modeling methods. The first step in block-wise ITSOEM building process, denoted as Block 0, is a null model meaning the IT- SOEM Block 0 model is built of constant only and no other independent variables are entered. The statistical parameters of successive block-wise logistic regression models are comparatively assessed to determine how the ITSOEM variables have performed within the block and from one block to another, compared with statistical outcomes recorded for the null model. The logistic regression models blockwise goodness-of-fit measures, as described by Hosmer et al., 9 are used to determine the validity of the and models. These goodness-of-fit measures include -2LL (ratio of likelihood of an event occurring determined from using the null model to that of likelihood of the same event determined using an alternative model), deviance (the difference between the two -2LL ratios), Hosmer and Lemshow Significance Test (evaluates the goodness-of-fit between predicted and observed probabilities), and Wald test of significance (similar to the t- statistic conceptually, and is a test of the null hypothesis that the model coefficient is equal to 0). Table 3. model goodness-of-fit measures for I-95 urban northbound 5 a.m. 1 p.m. analysis scenario. Goodness-of- Fit Measures Block0 Block1 I-95 Northbound Urban Segment: 5 a.m. 1 p.m. Analysis Scenario and Model Estimates The model block-wise goodness-of-fit measures estimated for the I-95 urban northbound (NB) segments from 5 a.m. to 1 p.m., as summarized in Table 3, confirm that including incident profile variables in Block1 and ITS detection variable in Block2 resulted in the reduced -2LL values. The reduction in -2LL values confirms that the Block1 and Block2 variables have improved model performance, compared with the null model Block0. The chi-square significance test for the deviance values also validate that the Block1 and Block2 variables serve as good model descriptors. However, a minute reduction of -2LL value and insignificant chisquare test for deviance value for Block3 indicates that the addition of Approach Volume and distance to DMS location as independent variables did not add any significant strength to the model. The Hosmer and Lemshow Significance Test results confirmed that the Block2 Block3 (Validated) -2LL (-2 log likelihood) Block-Wise Deviance Block-Wise Chi Sq. Significance H and L Significance Note: Block 0 is constant only, Block 1 is incident profile variables only, Block 2 is incident profile and detection system variables only, Block 3 is all model variables including incident profile, detection system, app. volume and distance to DMS. The validated has incident profile and detection system variables with interaction. H and L = Hosmer and Lemshow. Explanatory Variable Coefficient Estimate Model Estimates Wald Significance (p-value) Odds Ratio Class Interval D CCTV D VSP-CAD I HIGH PROFILE I MAJOR PROFILE D DMS D CCTV *D DMS Constant model variables exhibited good predictability, hence rejecting the null hypotheses for this test that the model variables are not capable of predicting the dependent variable. As a result, the models goodness-of-fit measures validate that the inclusion of incident profile and ITS detection variables into models consistently improve the performance of models. Additional model runs were performed to verify for any possible interaction between the model independent variables. The detection system independent variable D CCTV and distance to DMS location variable D DMS exhibited interaction and met the significance test. For the validated and interaction verified model, the coefficient estimates, Wald significance test p values, odds ratios, and class interval for odds ratios are summarized in Table 3. The Wald significance test p values for the model coefficients, with the exception of D VSPCAD variable, confirm that the model coefficients are significant at 0.05 p value and can predict the model outcomes effectively. The confidence interval for the odds ITE JOURNAL / MARCH

5 Goodness of Fit Measure Table 4. model goodness-of-fit measures for I-95 urban northbound during 5 a.m. 1 p.m. scenario. Block0 Block1 ratios estimated for these model coefficients are also observed to be within the acceptable limits. The D VSP-CAD variable at a p value of (an acceptable value is at or below 0.05) with odds ratio of and at a class interval of to (with a value of 1.00 being in the range) is the only variable that has exhibited poor significance. However, because of the D VSPCAD variable significance to the study and its contribution to overall explanatory strength, as confirmed in block-wise model building significance tests, this variable is included in the final effects model. Based on the above goodness-of-fit analysis and the model coefficient of estimates Wald significance tests, the final effects Block2 model for I-95 NB Urban Segment and for 5 a.m. to 1 p.m. analysis scenario is: Block3 Block4 (Validated) -2LL Block-Wise Deviance Block-Wise Chi Sq Significance H and L Significance Note: Block 0 is constant only; Block 1 is incident profile variables only; Block 2 is incident profile and DMS=0 or DMS=1 only; Block 3 is incident profile, DMS=0 or DMS=1, and incident duration; and Block 4 all model variables including app. volume and distance to DMS. The validated is incident profile, DMS=0 or DMS =1, and incident duration. (No interaction was found and the goodness-of-fit measures for DMS=0 and DMS=1 are identical.) Explanatory Variable Model Estimates with DMS = 0 Condition Coefficient Estimate Wald Significance (p-value) Odds Ratio Class Interval DMS =0 condition I HIGH PROFILE I MAJOR PROFILE Constant Model with DMS = 1 Condition DMS =1 condition I HIGH PROFILE I MAJOR PROFILE Constant log ( 1 ) DMS=0/1 = D CCTV D VSPCAD I HIGH I MAJOR D DMS * D CCTV * D DMS For, the model runs were performed for two distinct DMS conditions: DMS = 0 condition, when the incident detection was not followed by message dissemination on DMS; and DMS = 1 condition, when the incident detection was followed by message dissemination on DMS. The model block-wise goodness-of-fit measures and model coefficient of estimates significance test results for I-95 Urban NB segments during 5 a.m. to 1 p.m. hours of operation are summarized in Table 3. Similar to the Models, the models goodnessof-fit measures validated that the inclusion of incident profile variables, DMS =0,1, and variables, into the model building process consistently improved the performance of models. However, a minute reduction in -2LL value for Block 4 indicated that the addition of approach volume and distance to DMS location as independent variables did not add any significant strength to the model. Additional model runs were performed to verify for any possible interaction between the model independent variables. The model variables did not exhibit any interaction. For the validated and interaction verified model, the coefficient estimates, Wald significance test p values, odds ratios, and class interval for odds ratios for are summarized in Table 4. The p values for the model coefficients confirm that the model coefficients are significant at 0.05 p value and are capable of predicting the model outcomes effectively. The confidence interval for the odds ratios estimated for these model coefficients are also observed to be within the acceptable limits. Based on the above goodness-of-fit analysis and final model coefficient estimates, the final effects of the model for I-95 NB urban segment and for the 5 a.m. to 1 p.m. analysis scenario is as follows: For DMS = 0 condition the model is: log ( 1 ) SecInc=0/1 = DMS = I HIGH I MAJOR and for DMS = 1 condition the IT- SOEM2 model is: log ( 1 ) SecInc=0/1 = DMS = I HIGH I MAJOR and Model Results and Interpretation The odds ratios from the model are presented in Figure 2a. This odds ratio is the measurable value of the influence of CCTV detection on DMS 44 ITE JOURNAL / MARCH 2013

6 30 D(VSP) D(CCTV) I(Major) I(High) Odds of Messages Dissmeinated on DMS Figure 2a. interpretation of results for I-95 northbound urban segment 5 a.m. 1 p.m. analysis scenario. % Occcurrence of Secondary aincident Occurrence exp (Iduration)*0.009 expressed as % occurrence Incident Duration Figure 3. interpretation of results, influence of incident duration on I-95 urban segments. DMS(Message = 1) DMS(Message = 0) I(Major) I(High) messages on the model s dependent variable secondary incidents. This shows that the likelihood of secondary incident occurrence on the NB urban segments of the study corridor in the study time period is 2.33 times more when the primary incidents are not detected and messages are not posted on DMS signs. The odds ratios estimated for DMS =1 condition (meaning that the incident detection resulted in successful message dissemination for the primary incidents) shows that the likelihood of secondary incident occurrence dropped to 0.42, confirming the significant influence that DMS messages have on the deterrence of secondary incidents in the scenario. The other model coefficients I HIGH and I MAJOR have comparable high odds ratios, confirming that the high and major profile incidents have the propensity to cause secondary incioperations. This shows that of the three detection systems (D CCTV, D VSPCAD, and D SSP ), the CCTV detections on the NB urban segments of the study corridor in the study time period have an increased likelihood of being associated with the successful message dissemination on DMS. The VSPCAD detections have less likelihood of being associated with the message dissemination on DMS. In the same manner, the odds ratios for the incident profile descriptors I HIGH and I MAJOR serve as their respective measurable values of the influence on DMS operations. The odds ratios from the model coefficients are presented in Figure 2b. The coefficient DMS =0 (meaning that there was no message posted for the primary incidents that resulted in secondary incidents) has a high odds ratio, indicating the staggering influence of lack of DMS Odds of Secondary Incidents Occurrence Figure 2b. interpretation of results for I-95 northbound urban segment 5 a.m. 1 p.m. analysis scenario (with I MINOR as reference variables) Figure 2. Interpretation of results for and. dents compared with the minor profile incidents. The model s only continuous variable is (incident duration). For continuous variables, the odds ratio is interpreted as an exponential increase of a unit of the independent variable that in turn increases the odds of the occurrence of the dependent variable. As shown in Figure 3, the interpretation of the odds ratio of for means that in the first minute, the likelihood of primary incident followed by secondary incident increases by or for the NB urban segments of the I-95 corridor. If the incident duration increases by 25 minutes, then the likelihood of a secondary incident increases by, amounting to a 25 percent possibility of a secondary incident occurrence. Conclusions The primary objective of this study was to develop and apply an analytical methodology to quantify the effectiveness of ITS operations during incidents that occur on highways. By adopting block-wise logistic regression methods and using VDOT-provided ITS operations and incident management data for the I-95 study corridor, this study developed two ITS operations evaluation models: models to quantify the effectiveness of ITS detection systems during incident management, and models to quantify the effectiveness of DMS messages in the deterrence of secondary incidents. These two models yielded statistically significant model descriptors, and the odds ratios for each of the model s coefficients enable the practitioners to quantify the effectiveness ITE JOURNAL / MARCH

7 of ITS operations in a realistic manner. The odds ratios from the models allow the user to quantify the impact of each detection system descriptor and incident profile descriptor and develop pragmatic inferences from the estimated model coefficients. This is one of the principal contributions from this study, and meets the study core objectives of quantitatively determining the operations effectiveness of DMS and detection systems. models provide quantitative estimates of secondary incident deterrence because of the DMS =1 condition in the study corridor. The ITSOEM model results illustrate that transportation systems operators can apply ITSOEMs to quantify to what extent the deployed ITS systems are effective during the incidents that occur on I-95 study corridor. For instance, the ITSOEM model results discussed in this paper show that, CCTV incident detections from the study corridor are 4.7 times more likely to result in messages dissemination on DMS, compared with other detection systems such as SSP and VASPCAD. Thus, the model will enable transportation system operators to comparatively and quantitatively justify the existing and future CCTV deployments on I-95 corridor. Similarly, DMS message dissemination and transportation system operators ability to lower the propensity of secondary incidents occurrence from 2.33 times for a no message dissemination condition to 0.46 times for a with message dissemination condition enables them to develop meaningful and reliable benefits assessments of existing and proposed DMS deployments. Additionally, the study presents the importance of the granular operations data on ITS operations and incident management, and presents a unique and easy to apply data synthesis methodology. For the I-95 study corridor, the granular operations data made it possible to determine and quantify which detection system is most effective for incident management operations. The data syntheses methods and the treatment of and IT- SOEM2 analysis variables as dichotomous and binary variables provide a complete framework for practicing engineers to effectively determine how the ITS are useful in corridor operations, quantitatively determine potential ITS benefits in incident management, and develop strategies to fill in the gaps in ITS deployments. n References 1. U.S. Department of Transportation ITS Deployment Statistics. itsdeployment.its.dot.gov 2. Murthy, Gummada. Dynamic Message Signs (DMS) Operations Evaluation Models A Literature Review Report. Tampa, FL, USA: Civil and Environmental Engineering Department, University of South Florida, Murthy, Gummada, Jian Lu, and R. Lakshminarayan. Development and Application of Dynamic Message Signs (DMS) Operations Evaluation Models. TRB Annual Meeting, Virginia Department of Transportation. Six Year Operations Improvement Plan. Years Washington, S., M. Karlaftis, and F. Mannering. Statistical and Econometric Methods for Transportation Data Analysis. Boca Raton, FL, USA: Chapman and Hall/CRC, Savolainen, P. and I. Ghosh. Examination of Factors Affecting Driver Injury Severity in Michigan s Single-vehicle Deer Crashes. Transportation Research Record , Baublys, Adolfas and Aldona Jarašūnienė. Statistical Probability Evaluation of Operating ITS. Transport Research Institute, Vilnius Gediminas Technical University, Munley, Cheryl-Allen, Janice Daniel, and Sunil Dhar. Logic Model for Rating Urban Bicycle Route Safety. Transportation Research Record 1878, Hosmer, David and Stanley Lemshow. Applied Logistic Regression, Second Edition. John Wiley and Sons, Inc., Sinha, K., S. Peeta, M. Sultan, K. Ponnuru, and N. Richards. Evaluation of the Impacts of Technologies of the Borman Expressway Network. West Lafayette, IN, USA: School of Civil Engineering, Purdue University, Pande, Anurag and Mohamed Abdel-Aty. Comprehensive Analyses of the Relationship Between real Time Traffic Surveillance Data and Rear-End Crashes on Freeway. Transportation Research Record 1953, Cottrell, Wayne D. Estimating the Probability of Freeway Congestion Recurrence. Transportation Research Record 1634, Virginia Department of Transportation. Traffic Operations Center (TOC) Manual. Richmond, VA: VDOT, JIAN JOHN LU, Ph.D., P.E. is a senior professor and lead, Transportation Group at the University of South Florida, Tampa, Florida. He has led several national and international projects in traffic engineering, highway safety, ITS, and traffic flow analyses. He earned a B.S. in electrical automation engineering from the University of Science and Technology in Beijing, an M.S., in electrical engineering and transportation from Tongji University in China and a Ph.D. in transportation, civil engineering from the University of Texas at Austin. RAJARAM LAKSHMINARAYAN, Ph.D. is adjunct faculty in the Mathematics & Statistics Department of Global Health at the University of South Florida, Tampa, Florida. He is actively involved in development, education, and training of a series of biostatistics courses including, pharmacokinetics, SAS programming, and clinical database design and data management, for programmers, data managers, biostatisticians, physicians and healthcare professionals. Rajaram also provides global consulting services in clinical trial management and data analysis. He earned his B.Sc. and M.Sc. from the University of Mysore, his M.S. from the New Jersey Institute of Technology, and his Ph.D. from the University of South Florida. GUMMADA MURTHY, Ph.D, P.E. is associate program director, operations programs for the American Association of State Highway and Transportation Officials (AASHTO). He is responsible for traffic engineering, transportation systems operations and maintenance and ITS, and transportation security and emergency management programs. Gummada has more than 25 years of experience in all phases of transportation planning and engineering with a focus in roadway operations and roadside infrastructure maintenance. He is a graduate and holds a Ph.D. in civil engineering with a concentration in ITS Operations from University of South Florida, Tampa, Florida. He is a licensed professional engineer in Virginia. 46 ITE JOURNAL / MARCH 2013