Estimating the Probability of Rare Events Occurring Using a Local Model Averaging

Area of Science:

• Statistics
• Biostatistics
• Machine Learning

Background:

• Logistic regression is widely used for binary data analysis.
• Standard logistic regression models exhibit bias when analyzing rare events.
• Inaccurate probability estimates for rare events limit their application in critical fields.

Purpose of the Study:

• To develop an accurate method for estimating the probability of rare events using logistic regression.
• To address the limitations of standard logistic regression in handling imbalanced datasets.
• To provide a reliable approach for predicting rare occurrences in statistical modeling.

Main Methods:

• Proposed a local model averaging procedure combined with a data perturbation technique.
• Utilized different information criteria to generate multiple probability estimates.
• Employed an approximately unbiased Kullback-Leibler loss estimator for model selection.

Main Results:

• The proposed local model averaging approach demonstrated effectiveness in simulations.
• The method provides more accurate probability estimates for rare events compared to standard methods.
• The Kullback-Leibler loss estimator successfully identified the best probability estimates.

Conclusions:

• The novel method offers an improved approach for estimating rare event probabilities in logistic regression.
• This technique enhances the reliability of predictions for infrequent outcomes.
• The approach was validated using simulations and illustrated with a necrotizing enterocolitis (NEC) dataset.