Alternative approach to identify active effects in supersaturated experiments

The lack of the number of observations raises difficulty in estimating all possible main effects, as well as interaction effects, in the analysis of supersaturated experiments because the number of effects exceed the experiment size. There are some available methods in the literature. However they mostly deal with main effects only. In fact, the existence of confounding between main and interaction effects could lead to a misconclusion that a main effect is active but the true one is the interaction that is confounding with. Our novel approach employing genetic algorithm could simultaneously select appropriate main and interaction effects by considering the heredity principle. We implemented the approach to some data available in literature and revealed some good and useful results. The approach works with binary chromosomes representation with some restrictions to include corresponded main effects whenever the interaction is in the model.The lack of the number of observations raises difficulty in estimating all possible main effects, as well as interaction effects, in the analysis of supersaturated experiments because the number of effects exceed the experiment size. There are some available methods in the literature. However they mostly deal with main effects only. In fact, the existence of confounding between main and interaction effects could lead to a misconclusion that a main effect is active but the true one is the interaction that is confounding with. Our novel approach employing genetic algorithm could simultaneously select appropriate main and interaction effects by considering the heredity principle. We implemented the approach to some data available in literature and revealed some good and useful results. The approach works with binary chromosomes representation with some restrictions to include corresponded main effects whenever the interaction is in the model.