Using Unpruned Decision Tree and Random Forest For Learning From Multi-Class Imbalanced Data

Abstract

Learning from multiclass imbalanced data presents a great challenge for classicalmachine learning algorithms.The functions used for learning purpose of classifiers are naturally biased towards majority classes and less importance is given to minority classes. This paper compares the unpruned models induced usingdecision tree and random forest algorithm using appropriate metrics that are suitable for multiclass imbalance problem such as F-measure, Precision, Recall and G-mean. The experiments have been performed using a python based machine learning tool Sci-Kit on multiple datasets imported from online benchmark UCI machine learning repository.