A study on unstable cuts and its application to sample selection

An unstable cuts-based sample selection (UCBSS) is proposed. The proposed method addresses problems associated with traditional sample selection methods based on distance calculation when compressing large datasets, that is, significant time requirements and computational complexity. The core idea of the proposed method is that the extreme value of the convex function will be obtained at the boundary point. The proposed method measures the boundary extent of samples by marking unstable cuts, counting the number of unstable cuts and setting a threshold, and then obtains unstable subsets. Experimental results show that the proposed method is suitable for compression of large datasets with high imbalance ratio. Compared to the traditional condensed nearest neighbour (CNN) method, the proposed method can obtain similar compression ratios and higher G-mean values on datasets with high imbalance ratio. When the discriminant function of the classifier is a convex function, the proposed method can obtain similar accuracy and higher compression ratios on datasets with significant noise. In addition, the run time of the proposed method shows obvious advantage.