Enhancing Peer-to-Peer Network Security using 3D Convolutional Autoencoder based Anomaly Detection

In this era, Peer-to-Peer (P2P) networks have developed as a revolutionized network for data distribution and alternate to traditional client-server network. The features of P2P networks allow a distributed group of users to stake resources in the form of digital content or information. However, P2P networks lack in controlling access and it has limited security. In this research, Three Dimensional-Convolutional Autoencoder (3D-CA) is proposed to improve the security in P2P network. Initially, the data is obtained from University of New South Wales Network Behavior 2015 (UNSW-NB15) dataset and then preprocessed with data cleaning to ensure data quality. After that, the P2P autoencoder architecture is employed to learn the normal patterns which employs the detection in P2P network. Finally, the proposed 3D-CA is trained to detect the anomalies based on reconstruction error between input node and reconstructed pattern. From the results, the proposed 3D-CA model attained better outcomes compared to existing K-means algorithm in terms of accuracy, precision as well as F1-score by achieving 0.764, 0.90 and 0.80 respectively.