A Novel Approach to Feature Encoding

Target based encoding is a feature encoding technique that replaces categorical values with the mean of the target variable for each category. This can be done for both classification and regression problems. In this paper, we propose a novel target encoding method that can be used for categorical as well as continuous data. We evaluate our novel approach, referred to as kappa (κ) encoding, in regression problems with a series of data types. The approach is robust and shows significant improvement over a series of Machine Learning Regressors. We demonstrate that our proposed method has the potential to improve the performance of Machine Learning models on a variety of regression tasks.