Show simple item record

dc.contributor.advisorCook, Diane J
dc.creatorSprint, Gina Lee
dc.date.accessioned2017-06-19T17:58:51Z
dc.date.available2017-06-19T17:58:51Z
dc.date.issued2016
dc.identifier.urihttp://hdl.handle.net/2376/12106
dc.descriptionThesis (Ph.D.), Computer Science, Washington State Universityen_US
dc.description.abstractAge, injury, or disease-related impairments can severely diminish one's ability to be mobile and perform everyday tasks. To detect changes in abilities, healthcare professionals administer standardized assessments. Sensor technology can be utilized to complement clinical assessments to gain a more objective and detailed view of functionality. Specifically for rehabilitation and everyday living situations, sensor technology is able to provide more information about mobility and reduce subjectivity in outcome measures. We hypothesize that data from sensors can be analyzed using machine learning techniques to provide insights on mobility changes related to rehabilitation and daily behavior. We validate this hypothesis by analyzing data from three diverse settings: wearable sensor data collected during rehabilitation, wearable sensor data collected during everyday living, and ambient sensor data collected from everyday living in smart home environments. To investigate mobility assessment for rehabilitation, we analyze wearable sensor data collected from rehabilitation patients as they perform a sequence of ambulatory tasks that closely resemble everyday activities. We present algorithms to process sensor signals, compute metrics that describe ambulation performance, and quantify changes in mobility over one week of rehabilitation. Furthermore, we train machine learning algorithms with sensor-derived features and clinical information as part of our Hybrid Clinical Sensor Prediction (HCSP) approach. We are able to achieve higher prediction accuracy with HCSP by including our wearable sensor-derived features. For analyzing changes in everyday mobility data, we formalize the problem of unsupervised physical activity change detection and address the problem with our Physical Activity Change Detection (PACD) approach. PACD is a framework that detects changes between time periods, determines significance of the detected changes, and analyzes the nature of the changes. We illustrate and evaluate PACD with physical activity data collected from older adults who participated in a health intervention study and data collected from ambient sensors embedded in smart home environments. Results indicate PACD detects several mobility changes in the datasets. The proposed algorithms and analysis methods are useful data mining techniques for unsupervised change detection with potential to track physical activity, detect behavior changes due to health events, and motivate progress toward health goals.en_US
dc.description.sponsorshipWashington State University, Computer Scienceen_US
dc.language.isoEnglish
dc.rightsIn copyright
dc.rightsPublicly accessible
dc.rightsopenAccess
dc.rights.urihttp://rightsstatements.org/vocab/InC/1.0/
dc.rights.urihttp://www.ndltd.org/standards/metadata
dc.rights.urihttp://purl.org/eprint/accessRights/OpenAccess
dc.subjectComputer scienceen_US
dc.subjectChange detectionen_US
dc.subjectData miningen_US
dc.subjectMachine learningen_US
dc.subjectMobility assessmenten_US
dc.subjectSensorsen_US
dc.subjectSmart environmentsen_US
dc.titleAnalysis of Changes in Sensor Data for Mobility Assessment
dc.typeElectronic Thesis or Dissertation


Files in this item

Thumbnail

This item appears in the following Collection(s)

Show simple item record