Smooth ROC curve estimation via Bernstein polynomials.

Area of Science:

• Statistics
• Biostatistics
• Medical Diagnostics

Background:

• The Receiver Operating Characteristic (ROC) curve is a standard tool for assessing diagnostic test accuracy.
• Accurate estimation of ROC curves is crucial for reliable classification of observations into two groups.

Purpose of the Study:

• To propose novel tuning parameters for ROC curve estimation using Bernstein polynomial smoothing.
• To develop an easily implementable ROC curve estimator with a naturally selected tuning parameter.

Main Methods:

• Bernstein polynomial smoothing applied to the empirical ROC curve.
• Analysis of real and simulated data sets under various distribution scenarios (symmetric, right-skewed).
• Extensive simulation studies to compare performance against existing estimators.

Main Results:

• The new estimator demonstrated uniform efficiency gains over the empirical ROC estimator.
• The proposed method showed competitive performance against eleven other smooth ROC estimators.
• Performance was evaluated using mean integrated square errors across diverse data distributions.

Conclusions:

• The novel Bernstein polynomial smoothing approach provides an efficient and practical method for ROC curve estimation.
• This estimator is a valuable tool for evaluating diagnostic test accuracy, particularly in complex data scenarios.
• The ease of implementation and strong empirical performance make it a competitive alternative to existing smooth ROC estimators.