In each winter term BiCDaS offers a lecture series on data science. It takes place on six different Fridays throughout the term.
We are proud to once again have convinced a highly skilled set of internal and external data science professionals to join us as speakers.
Knowing the travel time for a given route is of importance for logistics applications, but also for individual travel, as is documented by the provision of predictions for example in Google maps. In both applications we do not only want to know the expected travel time but also the associated uncertainty as typically being late might incur a larger penalty than being too early.
For predicting travel times a large range of different time series methods are applied using a very diverse landscape of data sources with associated strengths and problems. The methods used include a large number of time series analysis techniques dealing with univariate and multivariate data sets, linear and non-linear models.
In this talk I will describe how some of the underlying problems are solved using insights from domain knowledge and statistical data analysis methods. The main underlying theme is that brute force purely data driven modelling does not work. Also only theory driven modelling typically is not sufficient. It is the combination of these two approaches that leads to success.
I will also hint at some of the current challenges, both technological and institutional. This will lead to my answers to the question: Are we there yet, can we provide reliable travel time predictions?