Ioannis Tsapakis

Inferring hybrid transportation modes from sparse GPS data using a moving window SVM classification

Understanding travel behaviour and travel demand is of constant importance to transportation communities and agencies in every country. Nowadays, attempts have been made to automatically infer transportation modes from positional data, such as the data collected by using GPS devices so that the cost in time and budget of conventional travel diary survey could be significantly reduced. Some limitations, however, exist in the literature, in aspects of data collection (sample size selected, duration of study, granularity of data), selection of variables (or combination of variables), and method of inference (the number of transportation modes to be used in the learning). This paper therefore, attempts to fully understand these aspects in the process of inference. We aim to solve a classification problem of GPS data into different transportation modes (car, walk, cycle, underground, train and bus). We first study the variables that could contribute positively to this classification, and statistically quantify their discriminatory power. We then introduce a novel approach to carry out this inference using a framework based on Support Vector Machines (SVMs) classification. The framework was tested using coarse-grained GPS data, which has been avoided in previous studies, achieving a promising accuracy of 88% with a Kappa statistic reflecting almost perfect agreement.

Discriminant Analysis for Assigning Short-Term Counts to Seasonal Adjustment Factor Groupings

The assignment of short-term counts to groupings of seasonal adjustment factors is the most critical step in the annual average daily traffic estimation process; this step is also extremely sensitive to error resulting from engineering judgment. In this study, discriminant analysis is examined, and several variable selection criteria are investigated to develop 12 assignment models. Continuous traffic volume data, obtained in the state of Ohio during 2005 and 2006, are used in the analysis. Seasonal adjustment factors are calculated with individual volumes of the two directions of travel as well as the total volume of a roadway segment. The results reveal that the best-performing directional volume–based model, which employs the Raos V algorithm, produces a mean absolute error (MAE) of 4.2%, which can be compared with errors reported in previous studies. An average decline in the MAE by 58% and in the standard deviation of the absolute error by 70% is estimated over the traditional roadway functional classification. In addition, time-of-day factors are slightly more effective in identifying similar patterns of short-term counts than when they are combined with the average daily traffic. When directional-specific factors are used instead of total volume–based seasonal adjustment factors, the improvement in the average MAE is approximately 41%. This conclusion is consistent with previous research findings and may result from the division of the data set by direction essentially doubling the sample size, which in turn increases the number of assignment options for a short-term count.

Effects of Tube Strikes on Journey Times in Transport Network of London

The goal of this study was to investigate the impact of five underground strikes on journey times in Londons transport network during 2009 and 2010. The main data source for this study was automatic number plate recognition cameras, which were installed on the entrances and exits of 670 travel links that covered the vast majority of the network and were equivalent to a total length of 1,740 km. The determination of spatio-temporal differences of strike effects between the first and the remaining strike days, the identification of changes in departure and arrival times, and the estimation of travel time delays within central, inner, and outer London, as well as between inbound and outbound traffic, were the main objectives of the study. The total travel time within the examined areas, the excess delay, and the corresponding percentage difference in journey times were the main performance measurements used. The most significant results showed that the second day of strikes resulted in significant delays as opposed to the first strike days. The peaks elongated by approximately 45 to 60 min, while the unique full-day strike had the highest percentage increase in travel times, especially during the evening period (74%). Central London was generally affected the most, especially during the morning peak, which experienced an average increase in travel times of 35%, while Central London also had the highest percentage of negatively affected links (80%). The inbound traffic experienced, on average, high delays during the morning peak; the outbound traffic yielded greater delays during the evening period.

A space-time delay neural network model for travel time prediction

Research on space-time modelling and forecasting has focused on integrating space-time autocorrelation into statistical models to increase the accuracy of forecasting. These models include space-time autoregressive integrated moving average (STARIMA) and its various extensions. However, they are inadequate for the cases when the correlation between data is dynamic and heterogeneous, such as traffic network data. The aim of the paper is to integrate spatial and temporal autocorrelations of road traffic network by developing a novel space-time delay neural network (STDNN) model that capture the autocorrelation locally and dynamically. Validation of the space-time delay neural network is carried out using real data from London road traffic network with 22 links by comparing benchmark models such as Naive, ARIMA, and STARIMA models. Study results show that STDNN outperforms the Naive, ARIMA, and STARIMA models in prediction accuracy and has considerable advantages in travel time prediction.

A Bayesian analysis of the effect of estimating annual average daily traffic for heavy-duty trucks using training and validation data-sets

The precise estimation of annual average daily traffic (AADT) is of significant importance worldwide for transportation agencies. This paper uses three modeling frameworks to predict the AADT for heavy-duty trucks. In total, 12 models are developed based on regression and Bayesian analysis using a training data-set. A separate validation data-set is used to compare the results from the 12 models, spanning the years 2005 through 2007 and taken from 67 continuous data recorders. Parameters of significance include roadway functional class, population density, and spatial location; five regional areas – northeast, northwest, central, southeast, and southwest – of the state of Ohio in the USA; and average daily truck traffic. The results show that a full Bayesian negative binomial model with a coefficient offset is the most efficient model framework for all four seasons of the year. This model is able to account for between 87% and 92% of the variability within the data-set.

How tube strikes affect macroscopic and link travel times in London

Abstract The purpose of this study was to investigate the impact of the five strikes on the London Underground (metro) rail system, which occurred in 2009 and 2010, on macroscopic and road link travel times. A consequence of these strikes was an increase in road traffic flows above usual levels. This provides an opportunity to observe the operation of the road network under unusually high flows. The first objective involves the examination of strike effects on inbound (IT) and outbound traffic (OT) within central, inner and outer London. Travel time data obtained from automatic number plate recognition cameras are used within the first part of the analysis. The second more detailed objective was to investigate in spatio-temporal effects on travel times on five road links. Correlation analyses and general linear models are developed using both traffic flow and travel time data. According to the results of the study, the morning IT had approximately twice as much delay as the OT. Central London experienced the highest delays, followed by inner and outer London. As would be expected, the unique full-day strike in 2009 yielded the worst impact on the network with the highest percentage increase in total travel time (60%) occurring during the morning peak in the IT in inner London. The results from the link-level analysis showed statistical significance amongst the examined links indicating heterogeneous effects from one link to another. It was also found that travel time changes may be more effectively captured through time-of-day terms compared to hourly traffic flows.

Accommodation of Saltwater Temporary Pipelines on the Roadside

Drilling and completing oil and gas wells, particularly when using horizontal and hydraulic fracturing techniques, requires enormous amounts of water. Generally, it is cheaper for the industry to move fluids by pipeline than by truck, hence the interest in using permanent and/or temporary pipelines to transport water in areas where oil and gas developments take place. This paper describes temporary pipeline installation and operation practices and how they impact roadside maintenance activities as well as offering guidelines on how to install and operate temporary pipelines. A GIS database of temporary pipeline locations was developed from permits issued by the Texas Department of Transportation between July 2011 and August 2016. General trends indicate that temporary pipelines are typically 3, 4, 8, or 10 inches in diameter. Operators tend to favor certain highway segments to install temporary pipelines within the right of way. When multiple temporary pipelines are installed on segments repeatedly, this can affect maintenance operations. Several trends were observed that necessitate the development of guidelines for temporary pipelines. Many temporary pipelines were placed away from the right of way line, which creates conflicts with maintenance operations and results in some temporary pipelines being in the clear zone. Many temporary pipelines are not anchored in place and roll into the bottom of ditches or do not maintain a uniform alignment, which affects roadside maintenance. The percentage reduction in overtopping flow rate due to installing temporary pipelines through culverts is also analyzed.

Use of Support Vector Machines to Assign Short-Term Counts to Seasonal Adjustment Factor Groups

The traditional method of estimating annual average daily traffic (AADT) involves several steps, with the most critical being the assignment process that involves allocating short-term counts to groups of seasonal adjustment factors (SAFs). The accuracy of AADT estimates highly depends on the assignment step, which is subject to errors stemming from human judgment. Support vector machines (SVMs), a supervised learning and statistical method, are employed to construct a series of assignment models that are compared with the traditional method and discriminant analysis (DA) models. Traffic volume data obtained from permanent traffic recorders are used to train and validate the models. The analysis is conducted with SAFs calculated for each direction of travel and separately for two-way traffic. The results reveal that the Gaussian kernel-based SVM model yields the lowest errors, improving AADT accuracy by 65% and decreasing the standard deviation of absolute percentage error by 73.7% over the traditional method. Another finding is that the assignment errors of the directional volume-based analysis are lower than those of the total volume-based analysis by 41.8%. The comparison between the two model parameters examined—the average daily traffic and the hourly factors—indicates that the combined use of both parameters in SVMs is more effective than when hourly factors are used alone. However, in the case of DA, the opposite results are obtained. A possible explanation is that the SVM kernels transfer data from the input space to a feature space and thus provide the ability to assign counts effectively with different types of data within the same model.