Abstract
Acknowledging that digital tools are widely used for human well-being monitoring and analysis, it is important to ensure that not only the decisions made by the underlying prediction model can be explained to the user, but also that the model itself is structured in a comprehensible way. In this work, we focus on describing how transparent predictive models for assessing eating behaviour can be built. We use ecological momentary assessment (ema) data of 135 adults collected throughout the day for 2 weeks via a mobile application. The data includes emotions, locations, types of craved and consumed food, etc. The transparent predictive models are built incrementally for every participant, allowing immediate personalization which subsequently can help design tailored and just-in-time interventions. We demonstrate that the prediction performance provided by the transparent models is comparable to that of the other tested classification algorithms. We summarize several ways in which the constructed models can be used during and after the data collection.keywordsecological momentary assessmenttransparent modelseating behaviourincremental learningpersonalization.
Original language | English |
---|---|
Title of host publication | Explainable AI Within the Digital Transformation and Cyber Physical Systems |
Subtitle of host publication | XAI Methods and Applications |
Editors | Moamar Sayed-Mouchaweh |
Publisher | Springer, Cham |
Chapter | 6 |
Pages | 91-124 |
Number of pages | 34 |
ISBN (Print) | 978-3-030-76408-1 |
DOIs | |
Publication status | Published - 2021 |