Optimizing Deep Feedforward Neural Networks for Effective Classification and Prediction of Cyber Incidents

This work starts with the Neural Network (NN) Feature Extraction (FE) stage and develops the Logarithm Neuron Layer (LNL) with a more robust data processing capability in order to address the limitations of Deep Learning (DL) and Machine Learning (ML) approaches for the prediction and classification of Significant Cyber Incidents (SCI). Since LNL operates logarithmic operations internally, unlike traditional neurons, it can analyze data without the need for data preprocessing phases. This reduces the influence of Data Normalization (DN) and other processes on the sample’s original data and increases the NN accuracy. Additionally, the initial values of logarithmic operations might be improved based on the particular demands of the task. Utilizing LNL in conjunction with a Deep NN (DNN), the Optimized Deep feed forward logarithmic neural network (ODFFLOGNN) was created on that base, where Honey Bader Algorithm (HBA) for optimal adjustment of the hyperparameters. The significance of cyber security and the effect of COVID-19 on cyber security are clarified by this research. Pre-pandemic and post-pandemic SCI are the two subgroups of the dataset, which is referred to as SCI in the Center for Strategic and International Studies (CSIS) report. Lastly, the research confirms that OFFLOGNN performs better than other methods for intrusion detection.