Cybersecurity in Public Space: Leveraging CNN and LSTM for Proactive Multivariate Time Series Classification

Mobile communications have become a vital domain for criminals and terrorists to exploit vulnerabilities, posing significant threats to public safety and national security. More precisely, they can employ cell site simulators such as International Mobile Subscriber Identity (IMSI) catchers to intercept and monitor mobile communications, enabling eavesdropping, tracking individuals’ movements, and potentially coordinating illegal activities while evading detection by law enforcement agencies. To overcome this issue, an innovative approach to detect IMSI catchers using Convolutional Neural Networks (CNN) and Long Short-Term Memory (LSTM) models is proposed. Leveraging the power of deep learning, the developed models process multivariate time series data to distinguish suspicious patterns indicative of IMSI catcher presence. This work compares nine different model architectures based on CNN, LSTM, or a combination of both through a series of case studies. We demonstrate that LSTM-based and Parallel CNNLSTM models outperform other architectures, achieving high precision and recall rates. Then, the best two models are tested with several sequence lengths. The presented models serve as valuable tools, providing a further enhancement to the security of mobile networks. The goal of this research work is to contribute to the broader mission of integrating artificial intelligence within the daily investigative practices of law enforcement agencies.