Feature Selection via Weighted Independent Domination

Many molecular biology data sets contain thousands of features, such as single-cell genes or omnic data. Standard predictive models perform poorly on such models because of the large number of features and correlations between them. The use of combinatorial methods, such as Independent Dominating Set (IDS) feature selection technique, perform well but seem to have been overlooked, perhaps due to their high computational complexity. We present an improvement to the IDS-based method that favors features with greater diversity and adopts the Weighted IDS problem as a model that is used as intermediate step in the feature selection process. Our approach is then tested on two genetic data sets, and the results are compared with those of previous methods. In addition to the notable improvements in the accuracy over [sic] traditional models, greater stability of the solution was observed.