Elegbeleye Femi Abiodun

Abstract

Cyber threats are evolving fast, and conventional means of cybersecurity-from rule-based intrusion detection systems to signature-based firewalls-have become ineffective against the new attacks. In this paper, we present a Hybrid Deep Learning Model, composed of CAE and LSTM networks, for the purpose of anomaly detection against cyber threats. CAE works toward the extraction of the relevant features by minimizing noise in the input while LSTM retains the sequential dependencies characteristic of network traffic for the effective detection of anomalies in a time series. The model was trained on the KDDCUP'99 dataset prepared with preprocessing using IQR-based normalization and Robust Scaling, alongside ACO for hyperparameter tuning. Experimental results validated the model performance, demonstrating an unprecedented accuracy of 98.97%, precision of 98.87%, recall of 98.67%, and F1 score of 98.75%, far above the performance offered by conventional models. The hybrid mod