Association Rules Mining on Multimodal Quantified-Self Data

Quantified-self refers to the practice that people use consumer wearables to collect personal data for self-care or self-knowledge. After more than a decade of its popularity, the quantifies-self practice is still mostly centered on data collection, while data analysis has been routinely limited to simple visualization and correlation analysis. In this study, we demonstrate how advanced data mining techniques could be leveraged to discover hidden associations that are difficult to spot through simple visualization. We applied association rules mining to discover interesting relations in a multimodal quantified-self dataset that contains self-tracking data from 10 participants. The data mining results convey actionable insights surrounding the associations between night sleep, physical activity, and glucose level. While some of the rules indicate intuitive associations between night sleep and physical activity the next day, we found surprising rules presenting the potential link between the day-time glucose patterns and the sleep structure in the previous night.