- Title Information
- Title
- Multi-Limb Sensor Fusion for Actigraphy-Based Sleep/Wake Classification
- Type of Resource (primo)
- dissertations
- Name:
Personal
- Name Part
- Green, Dylan Nathan
- Role
- Role Term:
Text
- creator
- Name:
Personal
- Name Part
- Gray, Marissa
- Role
- Role Term:
Text
- Advisor
- Name:
Personal
- Name Part
- Webb, Andrea
- Role
- Role Term:
Text
- Advisor
- Name:
Personal
- Name Part
- Hughes, Christopher
- Role
- Role Term:
Text
- Reader
- Name:
Corporate
- Name Part
- Brown University. Department of Computer Science
- Role
- Role Term:
Text
- sponsor
- Origin Information
- Copyright Date
- 2026
- Physical Description
- Extent
- 2, 35 p.
- digitalOrigin
- born digital
- Note:
thesis
- Thesis (Sc. M.)--Brown University, 2026
- Genre (aat)
- theses
- Abstract
- Actigraphy-based sleep monitoring has gained popularity due to its affordability
and unobtrusive nature compared to polysomnography (PSG), the clinical
gold standard. However, existing single-sensor actigraphy methods suffer from
a fundamental class imbalance; since the dominant class, sleep, constitutes
80 − 90% of a night of study, algorithms that optimize for sleep detection
accuracy consistently misclassify the minority wake class, inflating estimates
of total sleep time while under-reporting wakefulness. We directly address
this challenge with adaptations of the Cole-Kripke and Sadeh sleep detection
algorithms, modified to support data fusion at the data, score, and decision
levels from four Axivity AX6 actigraphy sensors, one on each wrist and ankle.
Rather than optimizing for sleep detection, we report F1, Cohen’s Kappa, and
MCC, evaluating our algorithms’ ability to detect both sleep and wake accurately.
Validated against simultaneous PSG at the Brown University Health Sleep Lab,
our results indicate that multi-sensor fusion eliminates placement-dependent F1
variance: the leading single-sensor algorithm had a 31% relative F1 difference
between wrists, a risk which every multi-sensor approach mitigates. As a
pilot study with four participants, we identified trends rather than specific
benchmarks: higher-level wake-biased fusion strategies consistently outperformed
lower-level fusion, thresholds from the seminal algorithms need to be adjusted for
multi-sensor approaches, and Cole-Kripke adaptations consistently outperformed
Sadeh. These findings establish multi-sensor actigraphy as a robust, low-cost
approach to wakefulness detection, with further implications for operational
contexts requiring simple, cheap monitoring hardware.
- Subject (fast)
(authorityURI="http://id.worldcat.org/fast", valueURI="http://id.worldcat.org/fast/01120819")
- Topic
- Sleep
- Subject (fast)
(authorityURI="http://id.worldcat.org/fast", valueURI="http://id.worldcat.org/fast/01029095")
- Topic
- Multisensor data fusion
- Subject
- Topic
- Class Imbalance
- Language
- Language Term (ISO639-2B)
- English
- Record Information
- Record Content Source (marcorg)
- RPB
- Record Creation Date
(encoding="iso8601")
- 20260516