Wavelet density estimation for mixing and size-biased data

Area of Science:

• Statistics
• Data Science
• Machine Learning

Background:

• Multivariate density estimation is crucial for understanding complex data distributions.
• Existing methods may not adequately handle mixed and size-biased data scenarios.
• Wavelet-based approaches offer a powerful tool for non-parametric function estimation.

Purpose of the Study:

• To develop and analyze wavelet estimators for multivariate density functions.
• To specifically address challenges posed by mixed and size-biased data.
• To derive theoretical guarantees for the performance of these estimators.

Main Methods:

• Utilizing wavelet decomposition for function approximation.
• Developing estimation procedures tailored for mixture models.
• Incorporating techniques to handle size-biased sampling.
• Deriving upper bounds for the Mean Integrated Squared Error (MISE).

Main Results:

• Established upper bounds for the MISE of wavelet estimators for the first time in this context.
• Demonstrated that the derived results align with existing theorems for independent and identically distributed data.
• Provided a theoretical foundation for using wavelet methods with complex data structures.

Conclusions:

• Wavelet estimation is effective for multivariate density functions even with mixed and size-biased data.
• The derived MISE bounds offer valuable insights into the performance and convergence rates of the estimators.
• This work extends the applicability of wavelet methods in statistical inference.