An Adaptive Synthetic Sample Coupled With Ensemble Multi-Filter Approaches For The High Dimensional Imbalanced Dataset
Abdulrauf Garba Sharifai, Ishola Dada Muraina, Usman Alhaji Abdurrahman · 2022
DNA microarray dataset infamously comprises a high dimensional and class imbalanced dataset, these problems pose severe challenges for the accurate identification of biomarker genes for cancer diagnosis and classification. The training learning model on the class imbalance dataset is a challenging task and became more tedious with high dimensional features. However, a small number of techniques have been proposed to address these issues simultaneously due to their complicated intersection. This paper proposes a hybrid approach of an adaptive synthetic sampling coupled with ensemble multi-filter for the high dimensional imbalanced datasets (ADASYN-EMF) method to overcome the problems of the high dimensional imbalanced datasets. An adaptive synthetic sampling approach for the imbalanced learning (ADASYN) method is used to generate the synthetic examples for positive class, while the ensemble multi-filter combined with Correlation-Based Redundancy is utilized to select the most robust gene subsets. Experiments on five DNA microarray datasets reveal that the proposed approach outperforms the state art methods. Thus, ADASYN-EMF is considered a robust technique that can handle the problem of high-dimensional imbalanced datasets.