On analysis of time-series data with preserved privacy

Time-series data analysis with privacy preservation is an open and challenging issue. To name a few are like analyzing company's confidential financial data, individual's health-related data, electricity consumption of individual's households and so on. Due to the complex nature of time-series data, analyzing such data without any revelation of sensitive information to adversaries is a pervasive task. Here, we have addressed the issue of analyzing numerical time-series of equal length with preserved privacy. Considering the Discrete Wavelet Transform as a suitable technique for transforming time-series in frequency-time representation, we have applied the concept in privacy-preserving analysis of such data. Experimental results show that our proposed method is superior to the existing methods in preserving the trade-off between data utility and privacy. The privacy models developed using the proposed method are also evaluated in terms of clustering and classification accuracies obtained from perturbed time-series data.