IceCube -- Neutrinos in Deep Ice The Top 3 Solutions from the Public Kaggle Competition

TL;DR

This paper analyzes the top solutions from a Kaggle competition focused on reconstructing neutrino directions in IceCube data, demonstrating state-of-the-art angular resolution achievements for cascade and track events.

Contribution

It provides detailed descriptions of the three best machine learning solutions, including data handling, architecture, training, and performance comparison.

Findings

Achieved angular resolution better than 5 degrees for cascade events above 10 TeV.

Achieved angular resolution better than 0.5 degrees for track events.

Top solutions outperform current state-of-the-art in neutrino event reconstruction.

Abstract

During the public Kaggle competition "IceCube -- Neutrinos in Deep Ice", thousands of reconstruction algorithms were created and submitted, aiming to estimate the direction of neutrino events recorded by the IceCube detector. Here we describe in detail the three ultimate best, award-winning solutions. The data handling, architecture, and training process of each of these machine learning models is laid out, followed up by an in-depth comparison of the performance on the kaggle datatset. We show that on cascade events in IceCube above 10 TeV, the best kaggle solution is able to achieve an angular resolution of better than 5 degrees, and for tracks correspondingly better than 0.5 degrees. These performance measures compare favourably to the current state-of-the-art in the field.