Comparative study on perturbation techniques in privacy preserving data mining

Abstract

Data Mining is a computational process that able to identify patterns, trends and behaviour from large datasets. With this advantages, data mining has been applied in many fields such as finance, healthcare, retail and so on. However, information disclosure become one of an issue during data mining process. Therefore, privacy protection is needed during data mining process which known as Privacy Preserving Data Mining (PPDM). There are several techniques available in PPDM and each of the techniques has its’ own benefits and drawbacks. In this research, perturbation technique is selected as privacy preserving technique. Perturbation technique is a method that alters the original data value before the application of data mining. In PPDM applications, perturbation technique able to provide a protection of data privacy but the accuracy of data should not be ignored too. In this research, three perturbation techniques are selected which are additive noise, data swapping and resample. For data mining techniques, two methods of classification are selected which are Naïve Bayes and Support Vector Machines (SVM). With the selection of these techniques, the experimental results are evaluated based on the hiding failure, accuracy and precision. For overall result, resample is selected as the best perturbation technique in naïve bayes and SVM classification for both glass and ionosphere datasets.