Tenfold Bootstrap as Resampling Method in Classification Problems

In this research, we propose the bootstrap procedure as a method for train/test splitting in machine learning algorithms for classification. We show that this resampling method can be a reliable alternative to cross validation and repeated random test/train splitting algorithms. The bootstrap procedure optimizes the classifier's performance by improving its accuracy and classification scores and by reducing computational time significantly. We also show that ten iterations of the bootstrap procedure are enough to achieve better performance of the classification algorithm. With these findings, we propose a solution to the problem of how to reduce computing time in large datasets, while introducing a new practical application of the bootstrap procedure.