Preliminary Assessment of an SFFS Method for Sub-Group Feature Identification in Heterogeneous Data Sets

Many biomedical pattern recognition problems involve disorders or conditions that present with different symptoms or features, resulting in a data set that is not homogeneous across an affected population. Examples of such data sets may include those describing autism spectrum disorders and mild cognitive impairment. In this paper, we describe preliminary analyses with synthetic data sets that simulate feature synergies inherent in many of these problems. A sequential forward floating search (SFFS) algorithm is then used to select relevant features for classification purposes. Our results suggest the SFFS method of feature selection may be an ideal technique when working with such data sets.