Supervised Discretization for Optimal Prediction
Wenxue Huang, Yuanyi Pan, Jianhong Wu · Procedia Computer Science · 2014
When data are high dimensional with a response variable categorical and explanatory variables mix-typed, a conveniently exe- cutable profile usually consists of categorical or categorized variables. This requires changing continuous variables to categorical variables. A supervised discretization algorithm for optimal prediction (with the GK-lambda) is proposed. The comparison of this algorithm with the supervised discretization for proportional prediction proposed in 1 is shown. Tests with some data sets from Machine Learning Repository(UCI) are presented.