Developing a Start-to-Finish Pipeline for Accelerometer-Based Activity Recognition Using Long Short-Term Memory Recurrent Neural Networks
Christian McDaniel
Shannon Quinn
Increased prevalence of smartphones and wearable devices has facilitated the collection of triaxial accelerometer data for numerous Human Activity Recognition (HAR) tasks. Concurrently, advances in the theory and implementation of long short-term memory (LSTM) recurrent neural networks (RNNs) has made it possible to process this data in its raw form, enabling on-device online analysis. In this two-part experiment, we have first amassed the results from thirty studies and reported their methods and key findings in a meta-analysis style review. We then used these findings to guide our development of a start-to-finish data analysis pipeline, which we implemented on a commonly used open-source dataset in a proof of concept fashion. The pipeline addresses the large disparities in model hyperparameter settings and ensures the avoidance of potential sources of data leakage that were identified in the literature. Our pipeline uses a heuristic-based algorithm to tune a baseline LSTM model over an expansive hyperparameter search space and trains the model on standardized windowed accelerometer signals alone. We find that we outperform other baseline models trained on this data and are able to compete with benchmark results from complex models trained on higher-dimensional data.
Neural Network, Human Activity Recognition, Recurrent Neural Network, Long Short-Term Memory, Accelerometer, Machine Learning, Data Analysis, Data Science, Hyperparameter Optimization, Hyperparameter
DOI10.25080/Majora-4af1f417-005