Explaining Deep Learning Models for Age-related Gait Classification based on time series acceleration
Authors:
** Zheng,
Bert Otten,
Michiel F Reneman,
Claudine JC Lamoth
Abstract:
Gait analysis holds significant importance in monitoring daily health, particularly among older adults. Advancements in sensor technology enable the capture of movement in real-life environments and generate big data. Machine learning, notably deep learning (DL), shows promise to use these big data in gait analysis. However, the inherent black-box nature of these models poses challenges for their…
▽ More
Gait analysis holds significant importance in monitoring daily health, particularly among older adults. Advancements in sensor technology enable the capture of movement in real-life environments and generate big data. Machine learning, notably deep learning (DL), shows promise to use these big data in gait analysis. However, the inherent black-box nature of these models poses challenges for their clinical application. This study aims to enhance transparency in DL-based gait classification for aged-related gait patterns using Explainable Artificial Intelligence, such as SHAP.
A total of 244 subjects, comprising 129 adults and 115 older adults (age>65), were included. They performed a 3-minute walking task while accelerometers were affixed to the lumbar segment L3. DL models, convolutional neural network (CNN) and gated recurrent unit (GRU), were trained using 1-stride and 8-stride accelerations, respectively, to classify adult and older adult groups. SHAP was employed to explain the models' predictions.
CNN achieved a satisfactory performance with an accuracy of 81.4% and an AUC of 0.89, and GRU demonstrated promising results with an accuracy of 84.5% and an AUC of 0.94. SHAP analysis revealed that both CNN and GRU assigned higher SHAP values to the data from vertical and walking directions, particularly emphasizing data around heel contact, spanning from the terminal swing to loading response phases. Furthermore, SHAP values indicated that GRU did not treat every stride equally.
CNN accurately distinguished between adults and older adults based on the characteristics of a single stride's data. GRU achieved accurate classification by considering the relationships and subtle differences between strides. In both models, data around heel contact emerged as most critical, suggesting differences in acceleration and deceleration patterns during walking between different age groups.
△ Less
Submitted 28 November, 2023; v1 submitted 20 November, 2023;
originally announced November 2023.
On-Device Bug Reporting for Android Applications
Authors:
Kevin Moran,
Richard Bonett,
Carlos Bernal-Cardenas,
Brendan Otten,
Daniel Park,
Denys Poshyvanyk
Abstract:
Bugs that surface in mobile applications can be difficult to reproduce and fix due to several confounding factors including the highly GUI-driven nature of mobile apps, varying contextual states, differing platform versions and device fragmentation. It is clear that developers need support in the form of automated tools that allow for more precise reporting of application defects in order to facil…
▽ More
Bugs that surface in mobile applications can be difficult to reproduce and fix due to several confounding factors including the highly GUI-driven nature of mobile apps, varying contextual states, differing platform versions and device fragmentation. It is clear that developers need support in the form of automated tools that allow for more precise reporting of application defects in order to facilitate more efficient and effective bug fixes. In this paper, we present a tool aimed at supporting application testers and developers in the process of On-Device Bug Reporting. Our tool, called ODBR, leverages the uiautomator framework and low-level event stream capture to offer support for recording and replaying a series of input gesture and sensor events that describe a bug in an Android application.
△ Less
Submitted 17 January, 2018;
originally announced January 2018.
