ExplainabilitySafetyTrustworthinesscs.HC
XAI-BayesHAR: A novel Framework for Human Activity Recognition with Integrated Uncertainty and Shapely Values
Anand Dubey, Niall Lyons, Avik Santra, Ashutosh Pandey
November 7, 2022
International Conference on Machine Learning and Applications
9 citations
1 influential
Key Takeaways
-
Uncertainty quantification improves real-world reliability: The framework tracks not just activity predictions but also statistical confidence levels in those predictions, which is critical for practical deployment in healthcare and smart home systems where incorrect classifications could have consequences.
-
The system can detect when it encounters unfamiliar situations: By using predictive uncertainty as an "out of distribution detector," XAI-BayesHAR can identify when sensor data falls outside its training experience—such as from sensor degradation or novel activities—and flag these cases rather than making unreliable guesses.
-
Explainability enables model efficiency: The use of Shapley values to identify which features matter most allows the model to be compressed by removing less important components, making the system more practical for deployment on resource-constrained devices like wearables while maintaining transparency about how decisions are made.
-
Sensor noise and degradation are addressed systematically: Rather than ignoring real-world problems like sensor wear and environmental variations, the Kalman filter component actively tracks and adapts to these changes over time, improving accuracy in actual deployment conditions.
Abstract
Human activity recognition (HAR) using IMU sensors, namely accelerometer and gyroscope, has several applications in smart homes, healthcare and human-machine interface systems. In practice, the IMU-based HAR system is expected to encounter variations in measurement due to sensor degradation, alien environment or sensor noise and will be subjected to unknown activities. In view of practical deployment of the solution, analysis of statistical confidence over the activity class score are important metrics. In this paper, we therefore propose XAI-BayesHAR, an integrated Bayesian framework, that improves the overall activity classification accuracy of IMU-based HAR solutions by recursively tracking the feature embedding vector and its associated uncertainty via Kalman filter. Additionally, XAI-BayesHAR acts as an out of data distribution (OOD) detector using the predictive uncertainty which help to evaluate and detect alien input data distribution. Furthermore, Shapley value-based performance of the proposed framework is also evaluated to understand the importance of the feature embedding vector and accordingly used for model compression