Skip to main content
. Author manuscript; available in PMC: 2019 Feb 26.
Published in final edited form as: Analyst. 2018 Feb 26;143(5):1147–1156. doi: 10.1039/c7an01888f

Fig. 4.

Fig. 4

Feature selection using mRMR vs. genetic algorithms. (a) The first iteration of mRMR selects an optimal feature testing all possibilities for maximum mutual information. Optimization of all features would be impractical. (b) Following iterations of mRMR are constrained by the initial feature(s) and therefore miss optimal values that deviate from this constrained subspace. Genetic algorithms introduce mutations that allow sampling outside of the subspace, reducing over-fitting.