Comparison of Decision Tree-Based Learning Algorithms Using Breast Cancer Data

Abstract

Recent times are witness to tremendous developments occurring in the field of Machine Learning. Data Mining applications are finding widespread acceptance, as they extend to encompass several sectors, including health care, education, weather forecasting, finance, etc. Data mining algorithms like classification algorithms are being integrated into the Knowledge Discovery in Databases (KDD) process to build optimized predictive models useful to healthcare and other professionals. Decision tree is considered to be one of the simplest and most widely used classifier algorithms. In the current study, we have compared six decision tree-based learning algorithms on breast cancer dataset. The algorithms used include, J48 decision tree, decision stump, random forest tree, REP tree, hoeffding tree and Logistic Model Tree (LMT). The dataset is obtained from Mizoram Cancer Institute, Aizawl, Mizoram, India and it contain 575 records and 24 attributes. Infogain attribute evaluator has been computed to determine the rank of an attribute. The analysis is performed in two steps. First step involves the building of the model by training data, while during the second step, the dataset is evaluated using 10-fold cross validation technique. Our computations indicate that of all the algorithms evaluated, J48 decision tree performs with the highest accuracy (84.2105%), while random tree demonstrates the lowest accuracy (76.4912%).