Discriminative classification models assume that both training and deployment data have same distributions of data attributes. These models give significantly varied performances when they are deployed under varied circumstances with different data distributions. This phenomenon is called Dataset Shift. In this paper we have provided a method which first determines whether there is a significant shift in the distributions of attributes between the training and deployment datasets. If there exists a shift in the data the proposed method then uses a Hill climbing approach to map this shift irrespective of its nature i.e. (linear or non-linear) to the equation for quadratic transformation. Experimental results on three real life datasets show strong performance gains achieved by the proposed method over previously established methods such as retraining and linear reframing.