Dynamic Fusion of LSTM Predictions Using Reinforcement Learning-Based GOWLA for Human Activity Recognition
Human Activity Recognition (HAR) contributes significantly to vital areas in healthcare, IoT, and smart monitoring applications. Generally, the current models rely on deep learning and traditional machine learning techniques such as LSTM and SVM. However, these methods face considerable challenges,...
Saved in:
| Main Authors: | , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
IEEE
2025-01-01
|
| Series: | IEEE Access |
| Subjects: | |
| Online Access: | https://ieeexplore.ieee.org/document/11036747/ |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Human Activity Recognition (HAR) contributes significantly to vital areas in healthcare, IoT, and smart monitoring applications. Generally, the current models rely on deep learning and traditional machine learning techniques such as LSTM and SVM. However, these methods face considerable challenges, such as imbalanced data distribution, weak adaptation of weights to temporal changes, and a decline in classification performance in rare activities. In this paper, we propose a new method based on dynamic reinforcement learning to update GOWLA weights, which is entitled Dynamic RL-GOWLA. In this method, the LSTM outputs are combined using Weighted Logarithmic Averaging with a reinforcement learning agent for dynamically updating the weights based on the model’s performance in each training iteration. Experiments on WISDM database show that the proposed method outperforms traditional models such as LSTM, SVM, Bayesian Classifier, RNN, and CNN, achieving a F1-score of 96.58%. This approach improves the adaptability of models to imbalanced data, making it a promising solution for tasks that require accurate classification of dynamic human activities. |
|---|---|
| ISSN: | 2169-3536 |