NegML: A Privacy-Preserving Machine Learning Approach Based on Negative Database

Abstract

Machine learning has become an increasingly prominent subject in the age of big data. It has made significant advances in image identification, object detection, and natural language processing, among other areas. The initial aim of machine learning is to extract meaningful information from enormous amounts of data, which unavoidably raises privacy concerns. Numerous privacy-preserving machine-learning approaches have been presented so far. However, most of them suffer from significant improvements in efficiency or accuracy. A negative database (NDB) is a data representation that may safeguard data privacy by storing and exploiting the complementary form of original data. In this research, we provide NegML, a privacy-preserving machine learning approach based on NDB. Private data are first transformed to NDB before being fed into machine learning algorithms such as a Multilayer perceptron (MLP), Logistic regression (LR), Gaussian naive Bayes (GNB), Decision tree (DT), as well as Random forest (RF). NegML has the same computational complexity as the original machine learning algorithms without privacy protection. Experiment findings on heart illnesses, milk datasets, Car evaluation benchmark datasets and Blood fusion dataset show that the accuracy of NegML is equivalent to the original machine learning model in most circumstances, as well as the technique based on differential privacy.