A Cluster-based Undersampling Technique for Multiclass Skewed Datasets

Imbalanced data classification is a demanding issue in mining and machine learning. Models that learn with imbalanced input generate feeble performance the minority class. Resampling methods can handle this balance skewed dataset. Cluster-based Undersampling (CUS) Near-Miss (NM) techniques are widely used However, these suffer from some serious flaws. CUS averts impact of distance factor on instances over majority Near-miss method discards inter-class within class elements. To overcome flaws, study has come up an undersampling technique called Adaptive K-means Clustering (AKCUS). The proposed blends clustering was analyzed aid experimental study. Three multiminority datasets different imbalance ratios were selected models created using K-Nearest Neighbor (kNN), Decision Tree (DT), Random Forest (RF) classifiers. results show AKCUS attain better efficacy than benchmark high ratios.