Feature selection is used to choose a subset of the most informative features in pattern identification based on machine learning methods. However, in many real-world applications such as online social networks, it is either impossible to acquire the entire feature set or to wait for the complete set of features before starting the feature selection process. To handle this issue, online streaming feature selection approaches have been recently proposed to provide a complementary algorithmic methodology by choosing the most informative features. Most of these methods suffer from challenges such as high computational cost, stability of the generated results and the size of the final features subset. In this paper, two novel feature selection methods called OSFSMI and OSFSMI-k are proposed to select the most informative features from online streaming features. The proposed methods employ mutual information concept in a streaming manner to evaluate correlation between features and also to assess the relevancy and redundancy of features in complex classification tasks. The proposed methods do not use any learning model in their search process, and thus can be classified as filter-based methods Several experiments are performed to compare the performance of the proposed algorithms with the state-of-the-art online streaming feature selection methods The reported results show that the proposed methods performs better than the others in most of the cases.
