Privacy-Preserving K-NN Classification Using Vector Operations

Abstract

This paper presents a privacy-preserving K-nearest neighbor (PPKNN) classification algorithm in privacy-preserving data mining (PPDM) domain to preserve privacy of customers in business organization. This paper is about modification of K-nearest neighbor (K-NN) classification algorithm using vector operations. It modifies each cell of the original data record by dividing into three sub-components as a single unit row vector. Similarly, the test data record is converted into cells of column unit vectors. Finally, the dot product is applied between row and column vectors to preserve the distance between the data records as original data records. In modified dataset when distances between the records are preserved, then PPKNN works similar to K-NN algorithm. In this work, PPKNN is applied on real datasets referred from UCI machine learning repository and compares classification accuracies with K-NN algorithm.