Perturbation Based Technique for Privacy Preserving Clustering of High Dimensional Data

Privacy of personal data is a fundamental human right. The freedom and transparency of data flow due to swift advances in data processing techniques and internet technology has heightened concerns of privacy. Reluctance to provide personal information could impede the success of data mining. Concern about the privacy of data is becoming an important concern in business, academic, defense and health care domains. Privacy-preserving data mining (PPDM) addresses these issues by striking a balance between privacy preservation and knowledge discovery. We propose a novel linear component analysis based transformation technique for Privacy preserving clustering to preserve the privacy of confidential data. We further evaluate the performance of this technique with the classical k-means clustering algorithm. The effectiveness of our new approach is demonstrated by various experiments conducted on synthetic data sets of varying dimensions. The accuracy of clustering has been computed before and after privacy preserving transformation using adjusted rand Index. Based on our results, we conclude that our method is an effective and feasible technique to build data mining models from perturbed data.