Transfer Learning for Error-Contaminated Poisson Regression Models.

Area of Science:

• Statistics
• Biostatistics
• Machine Learning

Background:

• Poisson regression is common for count data.
• Transfer learning can leverage source data to improve original data estimation.
• Measurement error and high-dimensionality are challenges in transfer learning for count data.

Purpose of the Study:

• To propose a novel strategy for handling error-prone count responses using transfer learning.
• To address measurement error and high-dimensionality in the context of transfer learning for count data.
• To improve prediction accuracy and mitigate model uncertainty through model averaging.

Main Methods:

• Developed a method to estimate parameters in measurement error models using source data.
• Employed transfer learning to derive a corrected estimator for count response variables.
• Integrated a model averaging strategy to enhance prediction and reduce uncertainty.

Main Results:

• The proposed method demonstrates satisfactory performance in simulations.
• The approach effectively handles measurement error in count data analysis.
• Validation using breast cancer data confirms the method's validity.

Conclusions:

• The novel transfer learning strategy successfully addresses measurement error and high-dimensionality in count regression.
• Model averaging further improves prediction and robustness.
• The method shows practical utility, as evidenced by breast cancer data analysis.