Hybrid Ensemble Model for Handling Class Imbalance Problem in Big Data Analytics
Peter Irungu Mwangi, Lawrence Nderu, Leah Mwaura Mutanu, Dorcas Gicuku Mwigereri · 2022 International Conference on Electrical, Computer and Energy Technologies (ICECET) · 2022
In big data analytics, class imbalance problem adversely affects the performance measure of machine learning classifiers as they tend to achieve higher accuracy by giving more attention to the instances from the majority class, thus ignoring those from the minority class, which are often the class of interest and of value. Techniques for handling class imbalance include; data level, algorithm level, and ensemble techniques. The ensemble models are seen to always perform better as compared to the other techniques. In our work we propose an ensemble model based on random oversampling and bagging models for handling class imbalance. Additionally, the genetic algorithm technique has been adopted for feature selection, and the random forest and decision trees are used as machine learning classifiers. The experimental results showed that our ROS-Bagging ensemble model outperformed ROS and bagging separately to handle class imbalance. There was steady rise in precision value for all the classifiers used when trained using a dataset balanced using ROSBagging ensemble model.