|
Activity Number:
|
240
- Distributions and Significance
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Monday, July 30, 2018 : 2:00 PM to 3:50 PM
|
|
Sponsor:
|
Biometrics Section
|
|
Abstract #328862
|
|
|
Title:
|
Comparison of Interval Estimation in Machine Learning
|
|
Author(s):
|
Dai Feng* and Andy Liaw and Vladimir Svetnik
|
|
Companies:
|
Merck and Merck & Co., Inc. and Merck
|
|
Keywords:
|
machine learning;
prediction;
confidence interval;
frequentist;
Bayesian;
coverage
|
|
Abstract:
|
Prediction of response is a central task in Machine Learning (ML). From a statistical point of view, the randomness in the training/test data and ML algorithms is inherent and creates uncertainty of prediction. Estimates of this uncertainty and the interval estimates in particular is an area of active research in Statistics and Machine Learning. In this study, we investigate some frequentist and Bayesian approaches for the interval estimation. We compare the coverage of different methods using both simulated and real data. We find that generally the Bayesian approach tends to provide coverage closer to the nominal. Furthermore, the accurate coverage is not offset by a wider length.
|
Authors who are presenting talks have a * after their name.
Back to the full JSM 2018 program
|
