Dengesiz sınıflandırma sorunlarına torbalama ve arttırma esaslı yeni bir yaklaşım

Classification algorithms are employed in a wide range of real-world problems such as obstacle detection, fraud detection, medical diagnosis, spam detection, speech recognition, image processing, intrusion detection, and so forth. However, it is not always an easy task to propose a legitimate classifier. For a classification task, there are numerous limitations of datasets. One of the most confronted limitations in real-world classification tasks is skewed class distribution, also called the class imbalance problem. When learning is employed in class imbalanced datasets without incorporating appropriate adjustments into the existing algorithms, minority classes are mostly misclassified. This study introduces a novel classification algorithm that outperforms previous studies on benchmark datasets used for the class imbalance problem. The presented novel algorithm, namely, BagBoost, involves aggregating modified bagging and modified boosting algorithms to increase the visibility of minority class instances. The state-of-the-art algorithms in the classification of imbalanced datasets are investigated. The results of the best existing algorithms are compared with the proposed algorithm using benchmark datasets. Results show that BagBoost is a better classifier than commonly used classification algorithms in the literature for benchmark datasets according to F-measure and G-mean scores.