A Hybrid Method for Density Power Divergence Minimization with Application to Robust Univariate Location and Scale...

Area of Science:

• Statistics
• Optimization
• Robust Statistics

Background:

• Outliers can significantly distort statistical parameter estimation.
• Existing robust methods like MCD have limitations.
• Efficient optimization is crucial for reliable estimation.

Purpose of the Study:

• To develop a globally convergent optimization method for minimum density power divergence estimation.
• To address parameter estimation in univariate Gaussian data contaminated with outliers.
• To enhance the efficiency and robustness of statistical estimation.

Main Methods:

• A hybrid optimization procedure combining Newton's method and gradient descent.
• Incorporation of Armijo's rule for step control to ensure global convergence.
• Development of a minimum density power divergence estimator.

Main Results:

• The proposed method achieves global convergence.
• Extensive simulations show improved efficiency over the Minimum Covariance Determinant (MCD) estimator.
• The method is effective across various breakdown point values.

Conclusions:

• The new optimization method provides a more efficient and robust approach to parameter estimation.
• This technique is valuable for analyzing real-world datasets with potential outliers.
• The findings advance robust statistical estimation techniques.