Decision making: data driven vs. expert consultation

To measure is to know? Transport data can show ‘what’, but experts are essential to show ‘why’. New forms of transport data have enormous potential but the author argues that we need to update our theories and models to fully take advantage of it and avoid misuse. We need to be able to answer ‘what to’ questions and not only ‘what if ‘.

01. Introduction

Conventional decision making in the Transport sector has been traditionally strongly supported by field/domain experts, dedicated tools and models, always having raw data as a basis for the analysis. The recurrent problem was the lack of data or the difficulty in collecting this data. The recent technological advances have increased the quantity (and quality) of available data significantly, especially when these have been supported by a behavioural change thanks to which individuals participate actively to the creation/collection of data (Waze and Open Street Maps are examples of users’ participation in the creation of large and reliable datasets). As a result, nowadays the amount of data that can be generated and collected is enormous and of high quality.

When public authorities are faced with the task of solving concrete problems, they usually approach the solution with a “what to” type of question, to which experts respond by defining a set of scenarios and using tools (models) for answering a set of “what if” type of questions.

02. Conventional decision making process

The currently used transportation models were initially developed in the early 1950’s, where the lack of technical capabilities for data collection created the necessity for developing methodologies for the estimation of various variables (e.g. travel time based on traffic flow using the BPR function). However, nowadays new data sources and big data are able to provide information of higher value, previously not available for supporting decision making. An interesting (and real-world) example of this is to create a park without sidewalks and let the people define those by using the park. The data collected in this case will be the marks of the paths followed by the people, which after a few weeks will define a clear set of routes within the park, defined by and for the users. Once the “data collection period” ends, the sidewalks can be built in the park following the user-driven design.

Indeed, during the last years, immense developments in the computer science domain have taken place. The continuous technological advances and a societal shift towards information sharing and gamification have resulted to large-sized datasets (big data) being collected and to some extent made available (open data), which grow exponentially.

In the Transport domain, big data includes significant pieces of information that, if correctly processed, are able to provide a better understanding of both the mobility needs and the traffic status, by enriching significantly the quantity and the quality of the data collected until today. They are mainly of disaggregated nature, encompassing information of individual actors of the mobility spectrum (i.e. travellers/drivers and vehicles) in real time at a very low cost in comparison to the conventional (mostly aggregated) traffic data collection methodologies. As a counterpart to the highest quality of disaggregated data, the data filtering and processing needs have increased significantly due to this increase in the granularity of the data. The lately introduced big data can be included in the transportation modelling framework in order to provide significant improvements or even a re-formulation of the conventional and largely used transportation models and tools, if not in a new definition and development of the transportation models. Conventional modeling was mostly based on isolated traffic flow measurements with heterogeneous coverage, but current available data sets such as Floating Car Data and Bluetooth devices detectors allows for more homogeneous coverage.

The new models will be strongly based on activity-based and/or agent-based modelling techniques, which have been largely developed during the last years and are able to provide high-quality results, but require significantly large and detailed datasets. The table below highlights the most important issues of conventional transportation modelling and the new capabilities enabled by big data.

A non-exhaustive list of new data sets presenting important potential for the transport sector is the following:

• Floating Car Data collected by professional fleets and individuals.
• Bluetooth devices detections.
• GSM related data.
• Social media data.

From the above datasets mobility and activity patterns can be inferred, which is the main fuel of any transportation model. In Thessaloniki the Hellenic Institute of Transport (HIT) of the Centre for Research and Technology Hellas (CERTH) is collecting most of them (only the GSM data is not available) and using it for understanding mobility and activity patterns as well as for providing mobility services. A few figures of how the data is used in Thessaloniki are presented below:

Floating Car Data

This data is collected from a fleet of 1.200 taxis and is used for calculating Origin-Destination matrices for taxi users as well as for estimating traffic status in real time.

Bluetooth devices detectors

The network of Bluetooth devices detectors is composed by 43 units detecting Bluetooth-enabled devices, which detections are used for calculating Origin-Destination matrices and travel time in real time.

Social media data

Data from social media networks (Facebook and Twitter) is collected in Thessaloniki for inferring activity patters of the population.

Big Data Europe

It is worth to mention that Thessaloniki is the pilot location for the demonstration related to the Societal Challenge of the Transport sector, which will allow for an accurate and multi-source based prediction of the traffic status in the city.