Time-frequency analysis and autoencoder approach for network traffic anomaly detection

Detection of anomalies in network traffic is critical to mitigating cyber threats. This study integrates continuous wavelet transform (CWT), discrete-time Fourier transform (DTFT), short-time Fourier transform (STFT), and autoencoders to identify anomalous network behaviour. It conducts time- frequency analysis of pre-processed network traffic data such as packet size and duration, extracting meaningful features fed into an autoencoder. Reconstruction error deviations indicate anomalies like spikes or irregular oscillations. • This hybrid approach demonstrates good scalability for the real-time implementation of cybersecurity measures. Further developments can be made in autoencoder architectures to achieve their full potential in large-scale systems. • The model is robust and scalable for real-time applications, achieving 95% detection accuracy by identifying 72 anomalies. • Obtained results indicate that the approach is feasible for deploying in practical cybersecurity applications. © 2025