Validation of Practicality for CSI Sensing Utilizing Machine Learning
Tomoya Tanaka, Ayumu Yabuki, Mizuki Funakoshi, Ryo Yonemoto
TL;DR
This study tackles the practicality of CSI-based human posture recognition by evaluating five ML models on WLAN Channel State Information collected in two indoor environments. By analyzing model performance as a function of training data volume and spatial configuration, the authors demonstrate high accuracy (≥$85\%$) within a fixed environment but substantial degradation (≈$30\%$) when generalizing to a different space, highlighting a critical generalization gap. The work compares Linear Discriminant Analysis, NB-SVM, Kernel SVM, Random Forest, and CNN (with a CNN input reshape from $ (5,30,3,3,2) $ to $ (batch,9,30,5) $ using amplitude-only CSI) and provides concrete data-splitting and evaluation protocols that quantify data-efficiency and environmental robustness. The findings emphasize the need for environment-aware adaptation, diverse training data, and improved sensing hardware or alternative sensing modalities (e.g., RADAR beamforming) to realize practical CSI-based sensing in real-world settings. Overall, the paper contributes a rigorous framework and clear evidence that spatial generalization remains a major hurdle for CSI-driven posture recognition, guiding future work toward robust, transferable sensing systems.
Abstract
In this study, we leveraged Channel State Information (CSI), commonly utilized in WLAN communication, as training data to develop and evaluate five distinct machine learning models for recognizing human postures: standing, sitting, and lying down. The models we employed were: (i) Linear Discriminant Analysis, (ii) Naive Bayes-Support Vector Machine, (iii) Kernel-Support Vector Machine, (iv) Random Forest, and (v) Deep Learning. We systematically analyzed how the accuracy of these models varied with different amounts of training data. Additionally, to assess their spatial generalization capabilities, we evaluated the models' performance in a setting distinct from the one used for data collection. The experimental findings indicated that while two models -- (ii) Naive Bayes-Support Vector Machine and (v) Deep Learning -- achieved 85% or more accuracy in the original setting, their accuracy dropped to approximately 30% when applied in a different environment. These results underscore that although CSI-based machine learning models can attain high accuracy within a consistent spatial structure, their performance diminishes considerably with changes in spatial conditions, highlighting a significant challenge in their generalization capabilities.