An Experimental Study on Decision Tree Classifier Using Discrete and Continuous Data

Classification is one of the fundamental tasks of pattern recognition, data mining, and big data analysis. It spans across the domain for classifying novel instances whose class labels are unknown prior to the development of model. Decision trees like ID3, C4.5, and other variants for the task of classification have been widely studied in pattern recognition and data mining. The reason is that decision tree classifier is simple to understand, and its performance has been comparable with many promising classifiers. Therefore, in this work, we have developed a two-phase method of decision tree classifier for classifying continuous and discrete data effectively. In phase one, our method examines the database, whether it is a continuous-valued or discrete-valued database. If it is a continuous-valued database, then the database is discretized in this phase. In the second phase, the classifier is built and then classifies an unknown instance. To measure the performance of these two phases, we have experimented on a few datasets from the University of California, Irvine (UCI) Machine Learning repository and one artificially created dataset. The experimental evidence shows that this two-phase method of constructing a decision tree to classify an unknown instance is effective in both continuous and discrete cases.