[-] Show simple item record

dc.contributor.advisorMirielli, Edwardeng
dc.contributor.authorWaage, Mattheweng
dc.date.issued2023eng
dc.date.submitted2023 Springeng
dc.description.abstractThis project's aim was to create a platform for personalized health data analysis, testing, and prediction, making it easier for ordinary people who are interested in N-of-1 trials to do their own self-experiments and take control of their mental and physical health. In these studies a single subject is observed and different interventions are systematically evaluated on them over time. These are typically longitudinal, occurring over weeks or months, with several rounds of treatments and evaluations in the form of a number of AB assignments. In these studies wearable technology, trackers, apps, sensors, and other IoT devices may be used to record information about the subject multiple times per day or week, if not constantly. In this study a singular self-experimenter collected data on themselves from several different sources such as mood questionnaires and a Fitbit wearable, among others. This data from the various sources was merged so that a variety of statistical methods could be performed. A few different modes of experimenting went into this study. One experiment tested the claim that spending 15 minutes per day writing in a gratitude journal had an effect on the subject's mood. This was achieved through a BABABA crossover phase design study, with each of the three phases being 28 days, for a total of 84 days in the experiment. The tests done for this experiment were the more traditional ANOVARM and ANCOVA, which were used to discover whether the intervention (B) phases were significantly different from the baseline (A), with relation to the subject's mood. Another test compared the claim that there was a difference in the subject's mood between the two groups of pre-experiment and during the experimental phase, through a Mann-Whitney U test. The last part of the study was a more complex machine learning (ML) pipeline that sought to predict the subject's mood based on over 3 years of daily collected data. The ML pipeline ingested the data, created several different ML models such as random forests and support vector machines, and compared which model was best at predicting the subject's mood. Feature importance was extracted from the best model through SHapley Additive exPlanations (SHAP), where the weight of the various feature effects on the target, in this case the subject's mood, was obtained. This notified the subject which behaviors had an effect on their mood. These different modes of experimenting were then compared, to see which was easier to implement or understand for future self-experimenters.eng
dc.description.bibrefIncludes bibliographical references.eng
dc.format.extentx, 89 pages : illustrations (color)eng
dc.identifier.urihttps://hdl.handle.net/10355/96189
dc.identifier.urihttps://doi.org/10.32469/10355/96189eng
dc.languageEnglisheng
dc.publisherUniversity of Missouri--Columbiaeng
dc.relation.ispartofcommunityUniversity of Missouri--Columbia. Graduate School. Theses and Dissertationseng
dc.titleN-of-1 : better living through self-experimentationeng
dc.typeThesiseng
thesis.degree.disciplineData Science and Analytics (MU)eng
thesis.degree.grantorUniversity of Missouri--Columbiaeng
thesis.degree.levelMasterseng
thesis.degree.nameM.S.eng


Files in this item

[PDF]

This item appears in the following Collection(s)

[-] Show simple item record