Optimizing Two-level Supersaturated Designs using Swarm Intelligence Techniques.

Area of Science:

• Statistics
• Experimental Design
• Optimization

Background:

• Supersaturated designs (SSDs) are crucial for efficient screening experiments with numerous factors.
• Selecting optimal SSDs becomes computationally challenging as the number of factors increases.
• Existing methods may lack efficiency in finding optimal designs.

Purpose of the Study:

• To develop an efficient algorithm for finding optimal SSDs using swarm intelligence.
• To address the discrete optimization problem in selecting factor level settings for SSDs.
• To demonstrate the algorithm's effectiveness using the E(s^2) criterion.

Main Methods:

• A novel algorithm based on swarm intelligence is proposed.
• The algorithm targets the discrete optimization problem of selecting factor level settings.
• The E(s^2) criterion is used as an illustrative example for optimization.

Main Results:

• The proposed algorithm finds E(s^2)-optimal SSDs that meet theoretical lower bounds.
• The algorithm demonstrates superior or comparable efficiency to the traditional CP exchange method.
• It shows potential for identifying D3-, D4-, and D5-optimal SSDs.

Conclusions:

• Swarm intelligence provides an effective approach for optimizing SSDs.
• The developed algorithm offers computational advantages and robust performance.
• This method enhances the search for optimal designs in high-dimensional screening experiments.