Variable selection is an important procedure to select relevant features from large datasets in optimization problems. The use of epistasis concepts becomes an alternative to assess the gene (variable) interdependence and select the most significative variables. This chapter describes the Epistasis-based Feature Selection Algorithm (EbFSA). Such implementation was recently proposed in a doctorate thesis from Computer Science. It has been applied to solve multivariate calibration problems with multiple linear regression and demonstrated superiority over traditional techniques by selecting the smallest number of variables as well as obtaining the best model predictive ability.
Lauro Cássio Martins de Paula. Epistasis-Based Feature Selection Algorithm. Methods in molecular biology (Clifton, N.J.). 2021;2212:37-44
PMID: 33733348
View Full Text