MQP CS/DS TS2 Analyze data from wearable devices (like FitBits)

Richards, Maceo

Student Work

MQP CS/DS TS2 Analyze data from wearable devices (like FitBits)

Público Deposited

The work described in this report outlines a process for exploring interpretability of deep neural network architectures designed for a specific domain application within time series classification. Work presented here is the continuation of an endeavor to explore pre-symptomatic pathogen exposure detection given multimodal time series physiology data. Previously, no studies have been conducted assessing interpretability of the algorithms being developed for this purpose. Exploring interpretability is ultimately the driving force of trustworthiness between end users and the artificial intelligence platforms they will operate. The topic of pre-symptomatic pathogen exposure detection is within the broader domain of time series classification, and challenges within this domain are outlined. Various contending methods for addressing these challenges are also discussed. This report primarily details an adaptation of LIME (Local Interpretable Model-agnostic Explanations) to time series classification called LIMESegment and its implementation in the task of pre-symptomatic pathogen exposure detection.

This report represents the work of one or more WPI undergraduate students submitted to the faculty as evidence of completion of a degree requirement. WPI routinely publishes these reports on its website without editorial or peer review.

Creator