Exploring Novel Change Point Analysis in Sports to Other Domains

Organisers: Dennis Reidsma, Katharina Proksch, and Rupsa Basu, University of Twente, the Netherlands.

Summary: The project “Sports, Data, and Interaction” (SDI) at the University of Twente (NL) explores the interplay between novel sensor measurement and processing approaches, novel mathematical (statistical) approaches for analysing the data in real time, and novel forms of interaction technology that use the data in order to offer added value during sports training.

One outcome of the SDI project is a statistical approach that allows you to analyse time series data (in the project: movement sensor data from runners) to find the statistically significant time point in the data where data has become “provably different”.  A runner’s data slowly changes over time in various ways, as the runner gets warm and settles into a rhythm, slowly gets more tired, and at some point reaches a point of exhaustion. Using our new method, we can not only point out the change point (where the runner gets tired to the point of starting to move differently), we can also make various other statements about the runner’s movement characteristics in relation to the exact changes that the method detects.

The proposed workshop will introduce the method to a wider audience beyond sports, and allow researchers and practitioners to try out the method for themselves and discuss its possible added value. Since the method is not dependent on the data domain, participants can come from any field.  

We expect three types of participants.

  • Those who want to apply the method to their own time series data set with possible change points as part of the workshop. These participants can contact the organisers preparatory to the workshop to discuss the precise content and format of the data so we can apply the method for them. During the workshop, these participants will briefly pitch their own dataset, followed by a presentation by the organisers of the results from applying the method, followed by a plenary discussion about possible interpretations of those results. Contact the organisers if you are interested.

Those who only want to discuss possible applicability of the method, but do not want to actually apply the method yet for the workshop. For these participants, we offer the possibility to quickly pitch their data set during the workshop, followed by a plenary discussion of how change point detection might apply to their work. Contact the organisers if you are interested.
•    Those who want to just come to the workshop and listen to explanations and results of how others worked with the method.  
 
Do you want to share your data set to have us apply the method to your data? Please contact us at d.reidsma@utwente.nl and we will discuss the precise format that will make your specific data set suitable for the workshop. Furthermore, we will discuss which features of the data are most likely to indicate change points, and we will apply the method for you (or help you apply it yourself).

What data sets are appropriate for this method? The method was developed for analysing time series data that contains change points. This can be on different time scales, e.g. human movement data from runners (as in our project), long-term observations with day / night cycle patterns, physiological data in which a stable state switches to a different state (e.g. fatigue, stress, …), etc. Please feel free to approach us when you are uncertain whether your data set is suitable. We are very interested to explore together with prospective participants what further kind of data sets lend themselves to being analysed with this method.

Who owns the results? Any results from the data to which the data is applied will in principle remain with the participant who supplied the data. The workshop leaders are at this point more interested in seeing the challenges and opportunities that participants see in working with the method.

Program outline:
•    Brief introduction to the project; presentation of some initial project results
•    Presentation of the new analysis method, illustrated through results obtained on our own data (sensor measurements of runners)
•    Presentation of results of the method on data supplied by participants: first, the data owner pitches the data set (5 min); followed by a presentation of the change point detection method results (5 min) with the organisers; followed by a plenary discussion about interpreting the results (5 min)
•    Pitches (5 min) from other participants people about their project, data set, and possible change points, followed by a plenary discussion (5 min) about the possibilities of using the new method
•    Plenary discussion between all participants about the pros, cons, opportunities and challenges of the novel method.

Change point analysis