A No Parameter Synthetic Minority Oversampling Technique Based on Finch for Imbalanced Data

The synthetic minority oversampling technique(SMOTE) has emerged as a significant approach to address class imbalance challenges in machine learning. However, the algorithm is afflicted by challenges such as the imbalanced distribution of minority class data and concerns regarding the quality of synthetic data. The enhanced variants combined with the clustering algorithm encounter the problems such as difficulty in determining the optimal value of hyperparameters and class overlap. So this paper proposes a new improved algorithm named NP-SMOTE. The core concept of the algorithm is as follows: initially, the FINCH algorithm is employed to cluster the minority class data into distinct clusters. Subsequently, the data within each cluster are categorized into boundary data and central data by determining the class of nearest neighbors for each minority class data. Finally, the appropriate synthesis methods are applied to generate data for these two classes of minority class data. This algorithm obviates the need for predetermined hyperparameters and circumvents the limitations of class overlap by synthesizing data from various classes in a customized manner. The algorithm exhibits robustness and superior generalizability as demonstrated by their comparison with commonly used algorithms across 6 datasets.