Parabolic Threshold Discretization For Big Data

The quality of knowledge extracted during the process of knowledge extraction from raw data is intrinsically dependent on the data pre-processing phase. Discretization is the fundamental function in the data pre-processing phase. The main objective of discretization is to tranform continuous infinite data into discrete finite data. Most of the discretization methods proposed in the literature are not adapted to Big Data context. In this paper, we propose a new discretization method that is adapted to Big Data. The method has been implemented and tested on several benchmark databases and also compared to well-known discretization methods. The results show that the proposed method is adapted to voluminous databases used in Big Data.