Robust estimation of unbalanced mixture models on samples with outliers

Area of Science:

• Statistics
• Machine Learning
• Data Science

Background:

• Mixture models are vital for representing heterogeneous data.
• Real-world data often contains outliers and unbalanced observations from sources.
• Robust estimation is crucial for accurate mixture modeling with contaminated, unbalanced data.

Purpose of the Study:

• To propose a novel robust mixture estimator.
• To effectively handle outliers and unbalanced mixture weights.
• To improve the estimation of mixture models in real-world scenarios.

Main Methods:

• Developed a robust mixture estimator incorporating outlier trimming.
• Utilized component-wise confidence level ordering for outlier identification.
• Validated the method against the FAST-TLE algorithm.

Main Results:

• The proposed method robustly estimates unbalanced mixtures across varying outlier fractions.
• Demonstrated superior performance on both synthetic and real-world neuroimaging data.
• Showcased applicability to samples where outlier fractions are initially unknown.

Conclusions:

• The novel robust estimator accurately models unbalanced mixtures with outliers.
• This method offers a reliable solution for real-world contaminated and unbalanced datasets.
• Applicable to diverse fields, including medical imaging analysis.