The cross-validated AUC for MCP-logistic regression with high-dimensional data.

Area of Science:

• Statistics
• Machine Learning
• Bioinformatics

Background:

• High-dimensional data presents challenges for traditional statistical models.
• Penalized methods are crucial for variable selection in sparse logistic regression.
• Existing criteria like AIC, BIC, and EBIC may not optimally balance model fit and prediction accuracy.

Purpose of the Study:

• To propose and evaluate a novel cross-validated area under the ROC curve (CV-AUC) criterion for tuning parameter selection.
• To integrate the CV-AUC criterion with the minimax concave penalty (MCP) for sparse, high-dimensional logistic regression.
• To enhance classification performance in binary outcome prediction.

Main Methods:

• Development of a CV-AUC criterion specifically for optimizing classification performance.
• Implementation of the minimax concave penalty (MCP) for variable selection.
• Derivation of an efficient coordinate descent algorithm for MCP-logistic regression.
• Comparative analysis with existing criteria (AIC, BIC, EBIC) through simulation studies.

Main Results:

• The CV-AUC criterion, when used with MCP, resulted in models with superior predictive AUC and lower classification error compared to AIC, BIC, and EBIC.
• Simulation studies demonstrated the finite sample performance of the proposed method.
• Application to microarray data for cancer-related gene identification showcased practical utility.

Conclusions:

• The CV-AUC criterion is an effective and attractive method for tuning parameter selection in penalized high-dimensional logistic regression.
• This approach improves classification performance, particularly in complex biological datasets.
• The proposed method offers a valuable alternative for researchers in bioinformatics and statistics.