# A novel approach to the bias-variance problem in bump hunting

## TL;DR

This paper introduces a new method for background-model selection in bump hunting that reduces bias and maintains sensitivity, providing reliable confidence intervals even with limited background knowledge.

## Contribution

A novel approach that simplifies bump hunting, minimizes bias in signal estimation, and ensures valid confidence intervals under minimal background assumptions.

## Key findings

- Reduces bias in signal-strength estimation
- Maintains sensitivity despite bias reduction
- Produces confidence intervals with valid coverage

## Abstract

This study explores various data-driven methods for performing background-model selection, and for assigning uncertainty on the signal-strength estimator that arises due to the choice of background model. The performance of these methods is evaluated in the context of several realistic example problems. Furthermore, a novel strategy is proposed that greatly simplifies the process of performing a bump hunt when little is assumed to be known about the background. This new approach is shown to greatly reduce the potential bias in the signal-strength estimator, without degrading the sensitivity by increasing the variance, and to produce confidence intervals with valid coverage properties.