Work

MissFit: towards mediating missing data in personal informatics (PI) systems

Year: 2024
Type: dissertation
Abstract: In today's fitness-focused society, individuals with diverse technical backgrounds are increasingly engaging in the collection and analysis of their fitness data through movements like the Quantified Self (QS) and dedicated communities focused on Personal Informatics (PI) systems. However, these tools face a common challenge—they often struggle to handle missing activity data effectively, thereby limiting users' ability to gain profound insights. Researchers in the Personal Informatics (PI) and Human-Computer Interaction (HCI) communities suggest that PI tools should ubiquitously collect data. Despite ongoing efforts to optimize battery life, a fundamental issue persists: users must actively remove the device for charging, posing a risk of forgetfulness, especially during vacations. This dissertation addresses the issue of missing data in Personal Informatics through four key contributions: 1) It explores existing algorithmic methods for handling missing data, constructs models for estimation, and evaluates their potential utility on a published Fitbit dataset. After this exploration, we discovered that the estimation models cannot make accurate estimates at an individual level, indicating a need for a new approach involving human-in-the-loop strategies. 2) We identify two distinct user groups and three primary usage behaviors from the semi-structured user study. These findings reveal that maintainers prefer to know the present, and trainees prefer understanding the past, and predicting the future. Missing data impacts trainees more than maintainers. Maintainers utilize data visualization and social features more on the PI tools, but trainees use data export and data analysis more. 3) We implement MissFit, a web app with three methods to handle missing data; these methods are algorithmic, event-based, and manual input; they are derived from natural approaches that fit with users' expectations discovered from the semi-structured user study. 4) After conducting an iterative user-centered design study, we identified four distinct user groups based on personalities and exercise routines: SW, SP, UW, and UP. People from different groups also present varying preferences for how to estimate missing data using different methods. All these contributions collectively lead to the proposal of six design implications. These implications offer design principles on how future PI tools can utilize users' experiences to guide algorithms in estimating missing data and how to balance efficiency and user experience.
Author: Wen, Nannan
Language: English
Open Access status: green
OpenAlex ID: W7030356506
Cites: 77
Cited by:
Related to: 10