Active Bagging Ensemble Selection

摘要

As technology progresses with more and more data collected, the need of finding the appropriate label for them increases. However, many times the labeling process is a very difficult or/and expensive task and in most cases a help of an expert or expensive equipment is needed. For this reason the need of labeling only the most appropriate instances rises. Active Learning techniques can accomplish this by querying only those instances that a trained model finds the greatest amount of information and providing them to a human expert in order to label them. Combining these techniques with a fast ensemble classifier, a very performant in terms of classification accuracy schema can emerge where a trained model in a small amount of labeled instances can grow by adding only the most informative instances from a much greater pool of unlabeled instances. In this paper, we will propose such a schema using Bagging Ensemble Selection that uses REPTree as base classifier under Active Learning techniques and we will compare it to four well-known ensemble classifiers under the same techniques on 61 real world datasets.