Title: Interaction-based variable selection approach for supersaturated design analysis
Authors: Ray-Bing Chen - National Cheng Kung University (Taiwan) [presenting]
Huei-Lun Siao - National Sun Yat-sen University (Taiwan)
Inchi Hu - Hong Kong University of Science and Technology (Hong Kong)
Shaw-Hwa Lo - Columbia University (United States)
Mong-Na Lo Huang - National Sun Yat-sen University (Taiwan)
Abstract: Supersaturated design is a well-organized design aiming at obtaining as much information as possible although while containing fewer factors involved. As it is not possible to estimate all effects in the experiment due to its size limitation, the main purpose of these types of designs is to discover influential factors under the factor sparsity assumption. We propose to identify the factors based on an influential measurement previously proposed. One major advantage of the new screening procedure is to be able to identify those factors with interaction effects influencing the experimental results. After influential factors and interaction effects are identified with fewer factors, we will apply the componentwise Gibbs sampler methodology to improve the accuracy of obtaining the exact set of significant factor effects. We will examine the effectiveness of the proposed two-stage analysis approach using simulations as well as four well-studied real examples in the literature. Finally, we compare our newly proposed method with others to examine the performances of screening factor effects.