Deep Learning Based Encryption Policy Intrusion Detection Using Commodity WiFi

WiFi-based intrusion detection plays a significant role in indoor safety applications, setting human free from wearable devices and causing no privacy concerns, compared to sensor-based or vision-based solutions. WiFi-based systems have achieved high accuracy, but with limitations in dataset collection consumptions and feature extractions. In this paper, we propose EPID, a scheme for Encryption Policy Intrusion Detection leveraging Channel State Information (CSI). EPID applies Butterworth low-pass filter for signal denoising, and conjugate calibration to remove CSI random phase offsets. Generative Adversarial Networks (GAN) based data augmentation approach is proposed to augment dataset. Besides, sparse autoencoder (SAE) is adopted for feature extraction to reduce computational complexity and mistaken detection risks caused by redundant statistics. Based on the features, one-class Support Vector Machine (SVM) is conducted for general intrusion detection. Extensive empirical evidence shows that EPID yields mean 96.6% detection accuracy with practical feasibility.