Unsupervised instance selection via conjectural hyperrectangles

Machine learning algorithms spend a lot of time processing data because they are not fast enough to commit huge data sets. Instance selection algorithms especially aim to tackle this trouble. However, even instance selection algorithms can suffer from it. We propose a new unsupervised instance selection algorithm based on conjectural hyper-rectangles. In this study, the proposed algorithm is compared with one conventional and four state-of-the-art instance selection algorithms by using fifty-five data sets from different domains. The experimental results demonstrate the supremacy of the proposed algorithm in terms of classification accuracy, reduction rate, and running time. The time and space complexities of the proposed algorithm are log-linear and linear, respectively. Furthermore, the proposed algorithm can obtain better results with an accuracy-reduction trade-off without decreasing reduction rates extremely. The source code of the proposed algorithm and the data sets are available at https://github.com/fatihaydin1/NIS for computational reproducibility.