Backpropagation-Based Deep Learning Model for Privacy-Preserving of Confidential Data

Methods for neural network learning that rely on backpropagation are commonly used in fields including intrusion detection, bioinformatics, homeland security, and medical diagnostics. These application sectors extract trends and patterns from enormous volumes of complex information. In the application categories mentioned, protecting personal information and sensitive data is a critical concern. Most of the backpropagation neural network learning techniques now in use do not consider how to safeguard sensitive data during learning. This study suggests using a backpropagation neural network technique since it preserves privacy and works well with horizontally partitioned data. The network weight vector must be produced for the training sample set when building the neural network. Information is not made public to preserve the neural network learning model's privacy. In this study, it is suggested that the weight vector is divided among all participants, with a portion of its value going to each participant. Each layer of the neural network safely computes the final and intermediate weight vector values of the network utilizing a secure multi-party computing protocol. The learning model developed may be safely distributed across all participants, who can then use it to predict the output corresponding to their objective data. The experimental results show that the recommended method outperforms traditional non-privacy-preserving algorithms in speed and accuracy.