Predicting model for multiclass imbalanced data using pipeline sampling technique with dynamic ensemble selection

Machine learning algorithms find patterns in data that helps to make better prediction and decision. In day to day life these algorithms help to make critical decision such as medical diagnosis, stock prediction fault detection etc., Classification algorithm predict labels from trained patterns. Imbalanced data classification distribute data unevenly among classes i.e majority class has high proportion data whereas minority class takes low proportion data. Common machine learning algorithms have poor prediction accuracy for minority class leads to data imbalance problem. Besides multi class imbalanced learning has greater challenges than binary classification. To address this issue, this works designed a new classifier model that combine pipeline sampling for resample the data and Dynamic ensemble classifier selection for prediction on multiclass imbalanced dataset taken form UCI repository. Performance are evaluated through the multiclass classification metrics such as Weighted average Accuracy, weighted average Precision and weighted average F1 score, Roc_AUC Score, cohen’s kappa score, Mathew correlation co-efficient. A thorough empirical comparison is conducted to analyze the performance of proposed model with existing ensemble algorithm Gradient Boosting Classifier Bagging classifier and Random forest classifier. Dynamic ensemble algorithm outperforms the existing ensemble algorithm