Pairing of most relevant variables and bootstrap samples with ridge regression for data sharing

A parameterized ridge regression-based perturbation model is proposed in this paper to address data privacy and data utility in cyber-physical systems. In this approach, most relevant confidential variables are extracted along with their bootstrap samples to attain high confidentiality protection and data classification accuracies. The multiple devices and their connectedness in cyber-physical systems can help the pairing of the most relevant confidential variables and their corresponding bootstrap samples efficiently. The out-of-bag error and signal-interference-ratio are used as measures to study the performance of the proposed model. The experimental analysis shows the existence of ridge regression parameters that can help perturb data to achieve high confidentiality and classification as data utility. This study motivates a further research with these models.