Providing feedback on cycling metrics in an adaptive training app

Abstract

Cycling technology has developed rapidly. With the advances of Artificial Intelligence and Machine learning, adaptive training applications have appeared. These provide personalised training plans that adapt to users’ characteristics such as weekly availability, goal and level. To be able to provide a sports coach-like experience to the user, developers want to imitate the process of sports coaching. One of the tasks of a coach is to provide feedback. Some adaptive training apps try to involve feedback in the form of data analysis and visualisation. However, actual scientific basis for the (design) decisions taken by these commercial applications is missing. In collaboration with the mobile adaptive training app JOIN, this thesis developed a way of providing post-hoc feedback on cycling metrics in a mobile application. We conducted focus groups to find out what kind of post-hoc feedback users of different levels of expertise want on training performed. Both experienced and less experienced users agreed on wanting feedback regarding the quality of specific intervals. Also, an accuracy score seemed to be the most logical way to show how well the planned training was performed. Lastly, they would like to receive more information regarding the way the training plan is adjusted after a performed workout. Using the focus group insights, we iteratively developed low-fidelity and high-fidelity prototypes. An implementation, the Workout Score, was chosen to be evaluated with a survey to assess how much users appreciate the post-hoc feedback on training performed. The survey showed that participants appreciated the Workout Score, finding it a simple and clear way to show the planned versus actual training. Yet, the explanation lacked specificity to better show where users need to improve to increase their score. Furthermore, the inclusion of RPE in the Score was unclear to some. Additionally, the Workout Score can motivate to stick to the suggested workout but may also demotivate when users receive a low score without actionable feedback. To improve the post-hoc feedback feature in the future, separate scores for each component of the Workout Score should be shown. To add to that, we suggest to reconsider the role of RPE or to provide a more comprehensive explanation for its inclusion. Lastly, we propose to make it more evident to users when they should select that they have completed the suggested workout and when it would instead be better to select that they have done something else. In the future, JOIN could explore if the algorithm could potentially take over this decision-making process.