Bagging with Randomised Low-Discrepancy Sequences

Abstract

A Bootstrap Aggregation (or Bagging for short), is a sample of a dataset with replacement. This means that a new dataset is created from a random sample of an existing dataset where a given row may be selected and added more than once to the sample. Consequently, like many randomised algorithms, most Bootstraps use pseudo-random number generators for their random decision making. Similarly, for the implementation of Monte Carlo Methods on computers, pseudo-random generators have been used to simulate the uniform distribution. The performance of the Monte Carlo Methods is known to be heavily dependant on the quality of the pseudo-random generators. In this paper, we investigate the randomised low-discrepancy sequences for Bagging. We experimented with the Bagging of the CART algorithm on some benchmark classification problems using randomised low-discrepancy sequences, and the results were compared with the same bagging using uniform initialisation with a pseudo-random generator. The results show that, Bagging with using randomised low-discrepancy sequences could help the Bootstrap Aggregation improve its performance.