Nirnayam

Abstract

Classification is a supervised learning process which requires the decision labels available for learning process. Decision classifier is a rule based approach for classification. This paper proposes a new approach for classification by proposing a new approach to build the decision tree. The construction of decision tree is based on the principle of distance computation instead of finding Gini index or Information Gain of split attributes. The distance function is the normalized Lp-norm distance function with p=2. The fundamental idea for obtaining decision tree is based on determining the class based probability vectors. The distance is computed for probability vectors computed for attributes and root node of the decision tree. The minimal distance attribute is chosen as the best choice and the dataset records are split based on the choice made. A working example that demonstrates the proposed approach for building decision tree is discussed for understanding the classification process.