Using Pairs of Data-Points to Define Splits for Decision Trees