Improved Sensitivity of the DRIFT-IId Directional Dark Matter Experiment using Machine Learning

Abstract

We demonstrate a new type of analysis for the DRIFT-IId directional dark matter detector using a machine learning algorithm called a Random Forest Classifier. The analysis labels events as signal or background based on a series of selection parameters, rather than solely applying hard cuts. The analysis efficiency is shown to be comparable to our previous result at high energy but with increased efficiency at lower energies. This leads to a projected sensitivity enhancement of one order of magnitude below a WIMP mass of 15 GeV c${}^{-2}$ and a projected sensitivity limit that reaches down to a WIMP mass of 9 GeV c${}^{-2}$, which is a first for a directionally sensitive dark matter detector.

References in corpus (12)

• A Unified Approach to the Classical Statistical Analysis of Small Signals
• Dark Matter Candidates from Particle Physics and Methods of Detection
• On the local dark matter density
• A New Model-Independent Method for Extracting Spin-Dependent Cross Section Limits from Dark Matter Searches
• Low Threshold Results and Limits from the DRIFT Directional Dark Matter Detector
• Old, Metal-Poor Extreme Velocity Stars in the Solar Neighborhood
• Measurement of the Range Component Directional Signature in a DRIFT-II Detector using 252Cf Neutrons
• Radon in the DRIFT-II directional dark matter TPC: emanation, detection and mitigation
• Reducing DRIFT Backgrounds with a Submicron Aluminized-Mylar Cathode
• Measurements of W-value, Mobility and Gas Gain in Electronegative Gaseous CS2 and CS2 Gas Mixtures
• Background Assay and Rejection in DRIFT
• Demonstration of ThGEM-Multiwire Hybrid Charge Readout for Directional Dark Matter Searches

Cited by in corpus (5)

• Improving ANAIS-112 sensitivity to DAMA/LIBRA signal with machine learning techniques
• The role of small scale experiments in the direct detection of dark matter
• Insights into Dark Matter Direct Detection Experiments: Decision Trees versus Deep Learning
• Progress in End-to-End Optimization of Detectors for Fundamental Physics with Differentiable Programming
• Bayesian technique to combine independently-trained Machine-Learning models applied to direct dark matter detection