BO-LCNN: butterfly optimization based lightweight convolutional neural network for remote data integrity auditing and data sanitizing model

• Published in: Telecommunication systems 2024-04, Vol.85 (4), p.623-647
• Main Authors: Flavia, B. Judy, Chelliah, Balika J.
• Format: Article
• Language: English
• Subjects: Algorithms; Artificial Intelligence; Artificial neural networks; Business and Management; Cloud computing; Computer Communication Networks; Cybersecurity; Data integrity; Data storage; Documents; Information storage; Integrity; IT in Business; Lightweight; Machine learning; Neural networks; Optimization; Pinholes; Privacy; Probability Theory and Stochastic Processes; Telecommunications systems
• ISSN: 1018-4864; 1572-9451
• DOI: 10.1007/s11235-023-01096-0

With the increasing use of cloud storage for sensitive and personal information, ensuring data security has become a top priority. It is important to prevent sensitive data from being identified by unauthorized users during the distribution of cloud files. The main aim is to transmit the data in a secured manner without encrypting the entire file. Hence a novel design for remote data integrity auditing and data sanitizing that enables users to access files without revealing sensitive information. Our approach includes identity-based shared data integrity auditing, which is performed using different zero-knowledge proof protocols such as ZK-SNARK and ZK-STARK. We also propose a pinhole-imaging-based learning butterfly optimization algorithm with a lightweight convolutional neural network (PILBOA-LCNN) technique for data sanitization and security. The LCNN is used to identify sensitive terms in the document and safeguard them to maintain confidentiality. In the proposed PILBOA-LCNN technique, key extraction is a critical task during data restoration and sanitization. The PILBOA algorithm is used for key optimization during data sanitization. We evaluate the performance of our proposed model in terms of privacy preservation and document sanitization using the UPC and bus user datasets. The experimentation results revealed that the proposed method enhanced recall, F-measure, and precision scores as 90%, 89%, and 92%. It also has a low computation time of 109.2 s and 113.5 s. Our experimental results demonstrate that our proposed model outperforms existing techniques and offers improved cloud data storage privacy and accessibility.