Comparison of parametric and non-parametric statistical features for Z-Wave fingerprinting

The number of internet connected devices by all accounts is set to increase dramatically in coming years as Internet of Things technologies become cheaper and more convenient. Z-Wave devices have found application in building control, smart energy, health care and equipment monitoring. Its closed standard ensures interoperability of devices and this stability has led to its popularity among consumers. As use of these devices becomes more widespread, the need to protect them becomes more important. In this research, the RF-DNA fingerprinting method is examined to protect these devices using their physical layer attributes. In particular, the traditional method of using parametric features such as variance, skewness, and kurtosis is challenged with the use of non-parametric features mean, median, mode and linear regression coefficient estimates. With careful analysis of variables, a 71% reduction in features is achieved while attaining >94% correct classification rate at an 8 dB lower SNR than using traditional parametric features.