Blind classification of e-scooter trips according to their relationship with public transport.

E-scooter services have multiplied worldwide as a form of urban transport. Their use has grown so quickly that policymakers and researchers still need to understand their interrelation with other transport modes. At present, e-scooter services are primarily seen as a first-and-last-mile solution for public transport. However, we demonstrate that 50% of e-scooter trips are either substituting it or covering areas with little public transportation infrastructure. To this end, we have developed a novel data-driven methodology that autonomously classifies e-scooter trips according to their relation to public transit. Instead of predefined design criteria, the blind nature of our approach extracts the city's intrinsic parameters from real data. We applied this methodology to Rome (Italy), and our findings reveal that e-scooters provide specific mobility solutions in areas with particular needs. Thus, we believe that the proposed methodology will contribute to the understanding of e-scooter services as part of shared urban mobility.

Analysis of Extended Information Provided by Bluetooth Traffic Monitoring Systems to Enhance Short-Term Level of Service Prediction.

Bluetooth monitoring systems (BTMS) have opened a new era in traffic sensing, providing a reliable, economical, and easy-to-deploy solution to uniquely identify vehicles. Raw data from BTMS have traditionally been used to calculate travel time and origin-destination matrices. However, we could extend this to include other information like the number of vehicles or their residence times. This information, together with their temporal components, can be applied to the complex task of forecasting traffic. Level of service (LOS) prediction has opened a novel research line that fulfills the need to anticipate future traffic states, based on a standard link-based variable, accepted for both researchers and practitioners. In this paper, we incorporate BTMS's extended variables and temporal information to an LOS classifier based on a Random Undersampling Boost algorithm, which is proven to efficiently respond to the data unbalance intrinsic to this problem. By using this approach, we achieve an overall recall of 87.2% for up to 15-min prediction horizons, reaching 96.6% predicting congestion, and improving the results for the intermediate traffic states, especially complex given their intrinsic instability. Additionally, we provide detailed analyses on the impact of temporal information on the LOS predictor's performance, observing improvements up to a separation of 50 min between last features and prediction horizons. Furthermore, we study the predictor importance resulting from the classifiers to highlight those features contributing the most to the final achievements.

Data-driven analysis of the impact of COVID-19 on Madrid's public transport during each phase of the pandemic.

COVID-19 has become a major global issue with large social-economic and health impacts, which led to important changes in people's behavior. One of these changes affected the way people use public transport. In this work we present a data-driven analysis of the impact of COVID-19 on public transport demand in the Community of Madrid, Spain, using data from ticket validations between February and September 2020. This period of time covers all stages of pandemic in Spain, including de-escalation phases. We find that ridership has dramatically decreased by 95% at the pandemic peak, recovering very slowly and reaching only half its pre-pandemic levels at the end of September. We analyze results for different transport modes, ticket types, and groups of users. Our work corroborates that low-income groups are the most reliant on public transportation, thus observing significantly lower decreases in their ridership during pandemic. This paper also shows different average daily patterns of public transit demand during each phase of the pandemic in Madrid. All these findings provide relevant information for transit agencies to design responses to an emergence situation like this pandemic, contributing to extend the global knowledge about COVID-19 impact on transport comparing results with other cities worldwide.

Data-Driven Analysis of Bicycle Sharing Systems as Public Transport Systems Based on a Trip Index Classification.

Bicycle Sharing Systems (BSSs) are exponentially increasing in the urban mobility sector. They are traditionally conceived as a last-mile complement to the public transport system. In this paper, we demonstrate that BSSs can be seen as a public transport system in their own right. To do so, we build a mathematical framework for the classification of BSS trips. Using trajectory information, we create the trip index, which characterizes the intrinsic purpose of the use of BSS as transport or leisure. The construction of the trip index required a specific analysis of the BSS shortest path, which cannot be directly calculated from the topology of the network given that cyclists can find shortcuts through traffic lights, pedestrian crossings, etc. to reduce the overall traveled distance. Adding a layer of complication to the problem, these shortcuts have a non-trivial existence in terms of being intermittent, or short lived. We applied the proposed methodology to empirical data from BiciMAD, the public BSS in Madrid (Spain). The obtained results show that the trip index correctly determines transport and leisure categories, which exhibit distinct statistical and operational features. Finally, we inferred the underlying BSS public transport network and show the fundamental trajectories traveled by users. Based on this analysis, we conclude that 90.60% of BiciMAD's use fall in the category of transport, which demonstrates our first statement.

An activity-dependent hierarchical clustering method for sensory organization.

Biological and artificial sensory systems share many features and functionalities in common. One shared challenge is the management setup and maintenance of sensory topological information. In the case of a massive artificial sensory receptor array, this is an extremely complex problem. Biological sensory receptor arrays, such as the visual or tactile system, face the same problem and have found excellent solutions by implementing processes of sensory organization. Not only can biological sensory organization initiate the topological data construction, it can deal with growing systems and repair damaged ones. Importantly, it can use the patterned activity of sensory receptors to extract topological relationships. Using inspiration from these biological processes, we propose an activity-dependent clustering method for organizing large arrays of artificial sensory receptors. We present an algorithm that proceeds hierarchically by building a quadtree description of sensory organization and possesses many qualities of its biological counterpart, namely it can operate autonomously, it uses the patterned activity of sensory receptors and it is capable of supporting growth and repair.