Technical Report UALR06-02: Using Active Learning with Integrated Feature Selection

In this paper we present two very popular aspects in supervised Machine Learning algorithms: feature selection and active learning paradigm. Feature selection algorithms are widely used in Machine Learning to reduce the feature space representing given data samples. Active learning is very popular supervised Machine Learning technique that has been effectively used in Text Classification tasks to reduce training time and achieve high accuracy with small labeling cost. Using feature selection integrated with active learning seems like a great idea as it combines faster learning with smaller feature space dimensionality. Though promising, we observe through various experiments that feature selection when integrated within active learning process yields inferior accuracy results. We believe that the changing Feature space representation hurts the very core of active learning strategy. We present the rationale behind poor accuracy and prove it empirically.