Intrusion Detection Methodology

Artificial Neural Networks (ANN)

Artificial Neural Networks (ANN) sometime have another names called neural networks or connectionist models. ANN consist of interconnected nodes arranged in layers, with each node performing a simple computation and transmitting signals to nodes in subsequent layers (Sun and Tang, 2021). It simulating the brain's electrical activity and connections between neurons. These networks consist of processing elements called neurons or perceptrons, organized in layers or vectors. Each neuron receives inputs from other neurons and calculates an output based on these inputs. The connections between neurons have different weights. Input data are multiplied by these weights, simulating the transmission of information within the network. Adjusting these weights can learn and improve their performance just like human experience. Activation functions enhancing its capacity to learn complex patterns from data. Techniques like regularization, dropout, and batch normalization are often employed to mitigate overfitting and improve general performance (Sun and Tang, 2021).

Recurrent Neural Networks (RNN)

Recurrent Neural Networks (RNNs) is designed to process sequential data, particularly effective for time series analysis and forecasting, and also speech recognition. RNNs have connections that form directed cycles, allowing them to maintain an internal state or memory of past inputs. This enables RNNs to capture temporary context within sequential data. The methodology of RNNs able to recurrently apply the same set of weights across multiple time steps, propagate information from previous states to future predictions. Long Short-Term Memory (LSTM) and Gated Recurrent Unit (GRU) architectures are different type of RNNs, equipped with mechanisms to selectively retain or discard information over long sequences, prevent the problem of vanishing gradient and more effective learning of long-range data process (C Yin et al., 2017).

Long Short-Term Memory (LSTM)

(Source: https://www.scaler.com/topics/deep-learning/lstm/) Basic RNNs have problem to retain information over long sequences due to the diminishing impact of gradients during back propagation, LSTMs incorporate specialized memory cells with self-connected recurrent paths and gating mechanisms which are capable of selectively retaining or discarding information over multiple time steps, maintain context and learn dependencies over extended sequences. The architecture comprises three gate types: input gates, forget gates, and output gates that modulate information flow through memory cells by dynamically adjusting gate activation based on input data and past states.

Gated Recurrent Unit (GRU)

(Source: https://www.linkedin.com/pulse/recurrent-neural-networks-rnn-gated-units-gru-long-short-robin-kalia) Similar to Long Short-Term Memory (LSTM), GRU also have special mechanisms to regulate the flow of data through the network layer. GRU simplifies the LSTM architecture merging the cell state and hidden state, providing a more simple structure with fewer parameters with more efficiently (Agarap, 2018).