Finding Forensic Artefacts in Long-Term Frequency Band Occupancy Measurements Using Statistics and Machine Learning

Wireless real-time communication between users is a key function in many types of businesses. With the emergence of digital systems to exchange data between users of the same spectrum, usage of the wireless spectrum is changing and increasing. Long-term frequency band occupancy measurements, carried out in accordance with the requirements of the International Telecommunication Union, can be used to measure and store informative values for further forensic investigation. In the existing literature, there is very limited research on using that information for a forensic investigation due to a lack of relevant datasets, examination methods and valuable artefacts. In this paper, we present a new approach to identify forensically sound deviations, often referred to as outliers, from using a monitored frequency band. We present the medcouple method for statistically detecting and classifying outliers. Furthermore, we created two datasets of long-term frequency band occupancy measurements that were used to evaluate our approach. We also evaluated our datasets with different machine learning techniques, which demonstrate that Random Forest has the highest classification accuracy and sensitivity to automatically detect outliers. These datasets will also be made publicly available for further research.