DP-Denclue: An outlier detection algorithm with differential privacy preservation

A key data preparation method is the Denclue-based outlier identification algorithm in data mining. Howev-er, the conventional Denclue algorithm does not provide priva-cy protection and has a low outlier identification accuracy. In this paper, we use differential privacy technology, applying the Laplace mechanism to the process of the Denclue algorithm to add noise to achieve privacy preservation. Unfortunately, the noise addition has a downward trend on the Denclue outlier detection algorithm's accuracy. Therefore, the information entropy weight distance, which weights the attributes and amplifies the influence of significant attributes on the outlier data, is used instead of the Euclidean distance in the Denclue algorithm to measure the distance between two data points. This makes it simpler for the algorithm to identify outlier data. A differential privacy-based Denclue outlier detection algorithm (DP-Denclue) is proposed. Experiments with the UCI dataset show that the DP-Denclue algorithm has excellent performance.