Evaluating BiLSTM and CNN+GRU Approaches for Human Activity Recognition Using WiFi CSI Data

2025-06-11

AI Summary

This work addresses non-contact human activity recognition (HAR) using WiFi channel state information (CSI), where model selection remains empirically driven despite significant variations in CSI data resolution. Method: We systematically compare bidirectional long short-term memory (BiLSTM) and CNN–GRU hybrid models on two benchmark datasets—UT-HAR (low-resolution CSI) and NTU-Fi HAR (high-resolution CSI)—and introduce the “data resolution–spatiotemporal feature coupling” perspective to formalize the interplay between CSI characteristics and deep learning architecture design. Contribution/Results: We establish a principled, CSI-aware model adaptation framework, revealing that CNN–GRU achieves superior performance on low-resolution UT-HAR (95.20% accuracy), whereas BiLSTM excels on high-resolution NTU-Fi HAR (92.05%). These results validate the necessity of co-optimizing model architecture with CSI data resolution and provide an interpretable, reusable theoretical foundation and practical guidance for architecture selection in CSI-based HAR.

Technology Category

• Cognitive Modeling & Cognitive Systems: Neural Spike Coding
• Machine Learning: Multimodal Learning
• Planning, Routing, and Scheduling: Activity and Plan Recognition

Application Category

• Graph Algorithms and Modeling for the Web: Graph neural networks and deep learning approaches for Web-related graphs
• Web Mining and Content Analysis: Models for Web evolution
• User Modeling, Personalization and Recommendation: On-Device user modeling, personalization, and recommendation

Abstract

This paper compares the performance of BiLSTM and CNN+GRU deep learning models for Human Activity Recognition (HAR) on two WiFi-based Channel State Information (CSI) datasets: UT-HAR and NTU-Fi HAR. The findings indicate that the CNN+GRU model has a higher accuracy on the UT-HAR dataset (95.20%) thanks to its ability to extract spatial features. In contrast, the BiLSTM model performs better on the high-resolution NTU-Fi HAR dataset (92.05%) by extracting long-term temporal dependencies more effectively. The findings strongly emphasize the critical role of dataset characteristics and preprocessing techniques in model performance improvement. We also show the real-world applicability of such models in applications like healthcare and intelligent home systems, highlighting their potential for unobtrusive activity recognition.

Problem

• Compare BiLSTM and CNN+GRU for WiFi-based activity recognition
• Evaluate model performance on UT-HAR and NTU-Fi HAR datasets
• Explore real-world applications in healthcare and smart homes

Innovation

• Compares BiLSTM and CNN+GRU for HAR
• Uses WiFi CSI data for activity recognition
• Highlights dataset impact on model performance