Case-based reasoning (CBR) is an approach to problem-solving used in research for sports science in the past years. CBR is an intelligent experience-based solution solving system explained as similar problems have similar solutions, and easily adapted to various fields. In this work, we use case-based reasoning for predicting best possible finish-times for speed skaters given various external conditions. With inspiration from related research in recommendation systems for other sports, we studied a system handling the factors affecting speed skating and retrieving the most sim- ilar races for further prediction. The CBR system was modeled with the open-source software myCBR Workbench and SDK. This software retrieves cases with a restful API provided by the SDK based on the local-global similarity principle also defined in myCBR Workbench. Looking at the results, we conclude that a CBR system like this is suitable for our problem statement. Speed skating offers multiple non-numeric features that can make a signifi- cant difference in the results. We tested two strategies for calculating new finish-times, where we found that the median strategy performed the most optimistic results, and mean strategy had less consistency. We experimented with two retrieval approaches where the use of non-personal-best times gave the most consistent results due to the knowledge base included more applicable cases than the season-best approach. A possible improvement upon our system is to implement the revise and retain process, so the CBR model use experience from solved cases and evaluates the non-numerical parameters.
© Copyright 2019 Veröffentlicht von Norwegian University of Science and Technology. Alle Rechte vorbehalten.