Sample Selection Based Active Learning for Imbalanced Data

The majority of learning systems don´t take in consideration real world data problem and consider that the training sets are perfect. However, in real world data, this hypothesis is not always true. In fact, real world data is characterized by many different problems like redundancy, incoherence or the big size of data. In this paper we focus on the problem of imbalance between class. Many solutions were proposed to resolve this problem like the use of re-sampling techniques. Unfortunately, these methods don´t achieve a high performance of learning. On the other hand, it was reported that sample selection techniques (SS) improves the accuracy of classical algorithms of classification by reducing the size of data. In this paper we propose to apply SS method on an imbalanced data in order to select the training sample for Active Learning classifier.