Phishcasting: Deep Learning for Time Series Forecasting of Phishing Attacks

摘要: Phishing attacks remain pervasive and continue to be a source of significant monetary loss, identity theft, malware. One the challenges is that in most organizational settings, detection paradigm inherently about identifying reacting threats real-time, as they are unfolding. As way complement these efforts with greater foresight, we introduce idea phishcasting — forecasting phishing threat levels weeks or months into future. Given attack volume time series data noisy devoid traditional seasonal cyclical trends, extend framework utilize multiple series, auxiliary information alternate representations. We also CoT-Net, flexible, end-to-end CNN-LSTM based deep learning method for complex series. CoT-Net uses embeddings uncover correlations between patterns within across industry sectors. Using publicly available test bed featuring organizations’ over time, find outperform state-of-the-art methods. By showing might possible practical, our work has important proactive implications cybersecurity.