A folding sum transformation for binary classification
Kangrok Oh, Kar‐Ann Toh · 2016
In this paper, we introduce a formulation for a folding sum transformation and then investigate into its impact on binary classification. The proposed folding sum transformation can reduce dimension of data without a training process. The least squares estimation and a full multivariate polynomial expansion are utilized to apply the folding sum transformation for binary classification. Twelve binary data sets from the UCI machine learning repository are utilized in our experimental study. Our results show that the folding sum transformation can either enhance or have comparable accuracy performance at a lower training and testing computational cost comparing with that without using the transformation.