Selecting the Best Subset of Features Using a Game-theoretic Approach: Applications in Information Systems

The curse of dimensionality is a major issue in datasets related to Information Systems (IS) because of the volume of data coming from smartphones, cameras, wireless sensory networks, social media, Internet search, etc. For such datasets applying a proper feature selection method can boost the performance of prediction or classification methods. While there are many feature selection techniques that can be used in the IS domain, the performance of them is problem-specific and they may not perform well on many datasets. Therefore, in this study, we address this issue by developing a novel method that employs ideas from the field of game theory. A computational study on real-life classification IS datasets shows that our proposed method outperforms or do as well as other benchmarks.