Using Conservative Estimation for Conditional Probability instead of Ignoring Infrequent Case

TL;DR

This paper introduces a conservative estimation method for conditional probability using confidence intervals, which outperforms traditional estimators in experimental evaluations.

Contribution

The paper proposes a novel conservative estimator based on confidence intervals for conditional probability, improving accuracy over existing methods.

Findings

Outperforms popular estimators in experiments

Provides more reliable probability estimates in sparse data scenarios

Demonstrates effectiveness across multiple datasets

Abstract

There are several estimators of conditional probability from observed frequencies of features. In this paper, we propose using the lower limit of confidence interval on posterior distribution determined by the observed frequencies to ascertain conditional probability. In our experiments, this method outperformed other popular estimators.