randomHAR: Improving Ensemble Deep Learners for Human Activity Recognition with Sensor Selection and Reinforcement Learning

TL;DR

randomHAR introduces an ensemble deep learning approach with sensor selection and reinforcement learning to enhance human activity recognition, effectively addressing data noise and variability challenges.

Contribution

It proposes a novel ensemble method that optimizes sensor data selection and model combination using reinforcement learning, improving HAR performance over existing methods.

Findings

Outperforms state-of-the-art ensembleLSTM on six HAR datasets.

Effectively handles noisy data and intra/inter-class variability.

Demonstrates significant accuracy improvements in HAR tasks.

Abstract

Deep learning has proven to be an effective approach in the field of Human activity recognition (HAR), outperforming other architectures that require manual feature engineering. Despite recent advancements, challenges inherent to HAR data, such as noisy data, intra-class variability and inter-class similarity, remain. To address these challenges, we propose an ensemble method, called randomHAR. The general idea behind randomHAR is training a series of deep learning models with the same architecture on randomly selected sensor data from the given dataset. Besides, an agent is trained with the reinforcement learning algorithm to identify the optimal subset of the trained models that are utilized for runtime prediction. In contrast to existing work, this approach optimizes the ensemble process rather than the architecture of the constituent models. To assess the performance of the approach, we compare it against two HAR algorithms, including the current state of the art, on six HAR benchmark datasets. The result of the experiment demonstrates that the proposed approach outperforms the state-of-the-art method, ensembleLSTM.