Performance Comparison of Five Different Neural Networks for Human Activity Recognition
Yuqi Zeng · 2024
With the rapid development of technology, Human Activity Recognition (HAR) technology is widely applied in many fields. The introduction of Deep learning has made HAR technology more mature. However, in practical applications, due to the diversity and complexity of the environment, different model exhibits marked differences in performance with different scenarios. Therefore, the key of achieving successful HAR application is selecting the most suitable model. To address this issue, this article proposes a method for constructing a Human Activity Recognition model comparison testing framework. Five representative classifiers, namely Convolutional Neural Network (CNN), Multi-Layer Perception Machine (MLP), Support Vector Machine (SVM), Bidirectional Long Short Term Memory Network(Bi-LSTM), and the Combination of Spatiotemporal Convolutional Neural Networks and Long Short Term Memory Networks (TD-CNN-LSTM) were selected for in-depth comparative research, By comparing the loss and accuracy of five classifiers through confusion matrix, TD-CNN-LSTM classifier has the best performance and the highest accuracy. This framework provides researchers with a unified and standardized evaluation system, helping them make wiser decisions in practical applications.