ETSINF – Prácticas en Empresa

El pasado jueves 3 de mayo a las 12:30 en el salón de actos dentro del ciclo Conferencias de empresa, se ofreció una charla sobre Tecnologías para la salud y el bienestar, del Instituto ITACA, TSB Tecnologías. En este evento se presentaron proyectos desarrollados por este grupo y se plantearon diferentes opciones de colaboración con este grupo de investigación.

Documento

El próximo 8 de mayo (martes) a les 12:00 en la sala de juntas del DSIC,   João Mendes, profesor de  la Universidade do Porto impartirá una charla titulada «Machine Learning Applications to Transport Planning»

******************
Abstract of the talk on «machine learning applications to transport planning»
—
Bus companies made, in the last 15 years, an important effort in the acquisition of information systems. Some of those systems aim to obtain information to better control the bus service they offer. Namely, they acquire GPS data that allow to know the actual service done by the buses.
In this talk it will be presented a set of different experiments, all of them aiming the use of such data in order to improve the planning and the control at bus companies:

1. Travel time prediction for the planning: the goal is to predict
several days in advance the travel time in order to better adjust
the service of the bus drivers to the service needs. This was done
using projection pursuit regression, support vector machines and
random forests. Experiments addressing each focusing task will be
described. Additionally, heterogeneous ensembles using dynamic
selection were also used.
2. Bus schedule validation: the goal is to evaluate whether the days
covered by each existing schedule plan is well done. We use
clustering with dynamic time warping as distance measure on each bus
route. Then we use an ensemble clustering approach to obtain a
consensual partition of the days. The rules that defines which days
should be covered by each schedule plan (i.e., each consensual
cluster) are obtained using RIPPER, a rule induction algorithm.
3. Bus bunching prevention: the goal is to detect the existence of
failures in the schedule plan that potentially causes bus bunching
occurrences. This is done through PrefixSpan, an algorithm for
mining sequences of events.