Filter versus wrapper feature selection based on problem landscape features

Feature selection is a complex problem used across many fields, such as computer vision and data mining. Feature selection algorithms extract a subset of features from a greater feature set which can improve algorithm accuracy by discarding features that are less significant in achieving the goal function. Current approaches are often computationally expensive, provide insignificant increases in predictor performance, and can lead to overfitting. This paper investigates the binary feature selection problem and the applicability of using filter and wrapper techniques guided by fitness landscape characteristics. It is shown that using filter methods are more appropriate for problems where the fitness does not provide sufficient information to guide search as needed by wrapper techniques.