Implementation of The Backpropagation Algorithm to Improve the Effectiveness of Artificial Neural Network Models in Classifying Flooding Attacks

Abstrak

Flooding attacks such as UDP flood, SYN flood, and ICMP flood can disrupt network stability, requiring an effective early detection system. This study aims to build a classification model using artificial neural networks (ANN) with the backpropagation method to distinguish between normal traffic and flooding attacks. Data was collected through simulation in VirtualBox with Kali Linux as the attacker and Windows 10 as the target, and captured using Wireshark. The results of training and testing both libraries showed differences in performance between the two libraries. The PyTorch model produced a prediction accuracy of 94% for normal networks and SYN floods, and 100% for UDP floods and ICMP floods, with a total accuracy of 97%. In contrast, the TensorFlow model achieved an accuracy of 77% for normal networks, 80% for UDP floods, 95% for SYN floods, and 100% for ICMP floods, with a total accuracy of 88%. The comparison of the two models shows that a simple Multi Layer Perceptron neural network with the backpropagation method using the PyTorch library is quite effective in classifying flooding attacks.