Comparison of Feature Selection Methods for High-Dimensional Small Datasets

Machine learning (ML) techniques are widely applied in healthcare, ranging from precision healthcare to the possibility of preventive care. A general problem in clinical studies is the combination of high dimensionality and low sample sizes, complicating the process of extracting significant indicators from the feature set. In this paper, we compare the quality of three feature selection methods when the number of samples is low compared to the number of features. The results are compared to those of earlier research that used an optimization process applied to recursive feature elimination. The two filter methods, Anova and chi-squared, were unable to produce stable results, whereas the embedded Lasso method showed more promising outcomes.