Privacy Preserving and Detection of EEG Signals Using Machine Learning Methods

Major requirements of data de-identification is preserving data privacy measures. This lies particularly with the medical domain, patients’ data is very sensitive to be kept with promising security and privacy methods. Homomorphic encryption is the approach which is employed popularly in privacy preserving techniques. Though HE faces practical limitations such as very high computational complexity, noise elimination responsibly at the bit level or the values level. An encoding scheme is proposed that shall enable the HE to operate on the scenarios related to floating values with arbitrary precision. The application of the scheme is tested on EEG signals. The direct way of fitting associating with fixed and deterministic stages is used to accomplish the supervised machine learning methodology, which is designed for training. The effects of the scheme and error on runtime are assessed by experiments using synthetic data with manageable computational time complexity and tested on numerical and text data.