• emergent
• gps
• hotspot
• machine learning
• mining
• stigmergy
• taxi

The aim of this thesis is to identify and characterise both frequent and infrequent patterns in crowd dynamics of a metropolis like New York. To this purpose, a stigmergy based framework was developed. The elements composing the framework mimic biological mechanisms present in nature, thus allowing collective behaviours to emerge. The emergent paradigm on which the developed framework relies allows to avoid the explicit modelling of the system.
Our approach exploits taxis' positional data to identify high-density areas (hotspots) within a city, hence allowing us to analyse how the hotspots' activity levels evolve over time.
The hotspots found in the first phase of the work are then used to sample the number of people being picked up or dropped off at any given time. This set of samples composes a signal that is then used to recognise reference activity patterns arising during the course of a day. This is carried out by means of a second framework that enhances the activity level signals, enabling the frequent patterns to emerge. By clustering those frequent patterns, we are able to identify exceptional patterns which do not fall in any of the clusters defined.
On the basis of the results of this research, it can be concluded that there is indeed a cyclic behaviour that can be observed in passenger activity in a big metropolis as New York. Although this metropolis is diverse and many events take place daily, the stigmergy-based approach has proven to be an outstanding method to enhance frequent patterns, allowing them to emerge spontaneously. Moreover, exploiting the frequent patterns detected, it is possible to infer whether a new input day falls outside the known patterns, being it either a novelty or an anomaly.