A Novel Privacy-Preserving Approach Using Optimized Deep Learning for Secure Data Mining

Abstract

Preservation of privacy involves the use of methods to protect sensitive data. Data mining is the derivation of various patterns and insights from big data using statistical and machine learning tools. A privacy-preserving data mining protocol follows a methodological system to ensure the safety of data encryption, improving key generation. The proposed system architecture offers a strong cloud-based platform for data encryption and retrieval. Data preservation is performed using Brakerski/Fan-Vercauteren (BFV), where data is encrypted with the help of a secret key and transformed with the help of a random matrix to increase security. The secret key is constructed using the Double Exponential Smoothing Secretary Bird Optimization Algorithm (DES-SBOA), combining the double exponential smoothing with the Secretary Bird Optimization Algorithm (SBOA). The encrypted data is stored safely in the cloud, ensuring that it will not be accessed by the wrong users, but can still be used to produce MLP outputs with an accuracy of 57.5%, a privacy of 39.8%, a utility of 98.5%, a fitness of 62.9%, and an execution time of 341.5s.