Abstract
With the extensive deployment of smart environments and IoT devices, WiFi sensing has proven its significant convenience and contact-free sensing capabilities in supporting a wide range of applications. However, designing a ubiquitous WiFi sensing system for diverse real-world scenarios presents a substantial dilemma, as the system performance deteriorates when the testing data diverges significantly from the training data due to domain variations. To address this dilemma, existing studies need extra efforts to develop new features or even retrain the original model under environmental variations. However, these approaches have not efficiently resolved the dilemma. In this study, we conduct a comprehensive study on the domain variation problem to make WiFi sensing robust and accurate in practical applications. Our definition of domains is comprehensive and includes environmental conditions, surrounding settings, user differences, user orientations, user's positions relative to WiFi sensors, and user participation time frames. We design a novel conformal prediction framework that quantifies the conformity (i.e., similarity) between the testing and training WiFi samples, then labels the testing samples with the most probable class(es). Unlike traditional conformal prediction which relies on data from a single domain, we develop a new statistical (Type I) approach to assess the conformity of the testing WiFi samples to individual training domains and aggregate the outcomes. To further improve the framework's generalization, we design a (Type II) fusion approach that utilizes the inter-relationships among domains for more accurate conformity quantification. Built upon these two methods, kernel density estimation-based and SVM-based methods are developed to compute the conformity scores for new testing samples to make conformal predictions. Extensive experiments, utilizing both self-collected and publicly available datasets show that our framework can improve prediction accuracies ranging from 20.1% to 77.1% in three of the most representative WiFi-based applications across six types of domain variations.
| Original language | English (US) |
|---|---|
| Pages (from-to) | 13912-13928 |
| Number of pages | 17 |
| Journal | IEEE Transactions on Mobile Computing |
| Volume | 25 |
| Issue number | 9 |
| DOIs | |
| State | Published - Sep 1 2026 |
| Externally published | Yes |
All Science Journal Classification (ASJC) codes
- Software
- Computer Networks and Communications
- Electrical and Electronic Engineering
Keywords
- Conformal prediction
- WiFi sensing
- deep learning
- domain variations
Fingerprint
Dive into the research topics of 'Solving the WiFi Sensing Dilemma in Reality Leveraging Conformal Prediction'. Together they form a unique fingerprint.Cite this
- APA
- Author
- BIBTEX
- Harvard
- Standard
- RIS
- Vancouver