Title: Research on privacy protection method based on deep reinforcement learning algorithm in data mining

Authors: Yan Cai; Rui Xue

Addresses: School of Computer and Artificial Intelligence, Henan Finance University, Zhengzhou Henan, 450046, China ' School of Computer and Artificial Intelligence, Henan Finance University, Zhengzhou Henan, 450046, China

Abstract: Protecting data privacy is a critical issue in information security. However, traditional methods often hinder data mining efficiency and accuracy. This study aims to balance data security and mining efficiency to improve accuracy while ensuring privacy. Using the deep reinforcement learning algorithm, the target-optimised deep Q-network (T-DQN) is proposed. Multiple standard network datasets are used for testing. The proposed algorithm achieves higher Bayesian network representation accuracy (7.2%-24.7%) compared to PrivBayes under different datasets. Revenue is also increased (12%-17% higher than Sarsa and 29%-32% higher than Q-learning). Weak regret value is lower (698-2,573 lower than Sarsa and 984-1,327 lower than Q-learning). The algorithm demonstrates good convergence, adaptability, and superior performance compared to other algorithms. It provides a reference for improving privacy protection efficiency in data mining.

Keywords: data mining; deep learning; privacy protection; DQN; Bayesian network; target-optimised deep Q-network; T-DQN.

DOI: 10.1504/IJCSYSE.2024.142764

International Journal of Computational Systems Engineering, 2024 Vol.8 No.3/4, pp.210 - 219

Received: 25 Apr 2023
Accepted: 11 Jun 2023
Published online: 21 Nov 2024 *

Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article