ℓ 1 $$ {\ell}_1 $$ -Penalized Multinomial Regression: Estimation, Inference, and Prediction, With an Application to...

Area of Science:

• Statistics
• Machine Learning
• Biostatistics

Background:

• High-dimensional multinomial regression models are crucial for analyzing complex categorical data.
• Statistical inference for these models, unlike logistic regression, remains less explored.
• Existing methods often lack robustness or comprehensive inference capabilities.

Purpose of the Study:

• To analyze estimation and prediction errors in contrast-based penalized multinomial regression.
• To extend debiasing methods for valid statistical inference in multinomial models.
• To assess the robustness of the proposed method under model misspecification and non-identical data distributions.

Main Methods:

• Developed a contrast-based penalized multinomial regression framework.
• Extended existing debiasing techniques to the multinomial setting.
• Incorporated methods for constructing confidence intervals and hypothesis tests.
• Evaluated performance via extensive simulations and a real-world dementia progression dataset.

Main Results:

• The debiased method provides valid confidence intervals and hypothesis tests for multinomial regression coefficients.
• The approach demonstrates robustness to violations of standard statistical assumptions.
• Identified significant predictors for different dementia subtypes using the debiased method.
• Simulation results confirm the superiority of the debiased method over existing inference techniques.

Conclusions:

• The proposed debiased penalized multinomial regression method enhances statistical inference in high-dimensional settings.
• This robust methodology is effective for identifying important predictors in complex biological data, such as dementia progression.
• The findings offer a valuable tool for researchers in statistics, machine learning, and biostatistics.