Pareto-Front Optimization of Variance-Added Expected Loss with Interrelated Qualities.

Area of Science:

• Industrial Engineering
• Statistical Quality Control
• Optimization Theory

Background:

• The bias-variance trade-off is a fundamental challenge in industrial quality optimization, impacting accuracy and uncertainty.
• Traditional methods often address bias and variance separately, leading to potentially suboptimal decisions and increased overall risk.
• Interrelated quality characteristics require integrated approaches to effectively manage both deviation from targets and system uncertainty.

Purpose of the Study:

• To propose a Pareto-front optimization framework for a variance-added expected loss function in quality optimization.
• To simultaneously capture deviation from targets (bias) and system uncertainty (variance) using a unified loss function.
• To enable flexible adjustment of trade-offs between bias and variance for more balanced and efficient optimization.

Main Methods:

• Integration of multivariate quadratic loss with a variance term to form a variance-added expected loss function.
• Development of a weighted formulation to flexibly adjust the trade-offs between bias and variance.
• Application of Pareto-front analysis to reveal trade-offs between expected loss and variance.

Main Results:

• The proposed framework simultaneously captures bias and variance, unlike sequential methods that can increase total risk.
• A weighted formulation allows for flexible adjustment of bias-variance trade-offs, leading to more balanced optimization.
• Pareto-front analysis effectively visualizes the trade-offs, enabling informed selection of optimal quality designs.

Conclusions:

• The proposed Pareto-front optimization framework offers a more balanced and efficient approach to quality optimization by integrating bias and variance.
• The method provides valuable insights into the bias-variance trade-offs, empowering users to select designs aligned with their specific needs.
• Validation through examples and a case study confirms the practical applicability and effectiveness of the proposed approach for interrelated quality characteristics.