[Paper Review] Developing a Machine Learning Algorithm-Based Classification Models for the Detection of High-Energy Gamma Particles

Emmanuel Dadzie, Kelvin Kwakye|arXiv (Cornell University)|Nov 17, 2021

Astrophysics and Cosmic Phenomena4 citations

TL;DR

This study develops and evaluates multiple machine learning classification models for detecting high-energy gamma particles using data from Cherenkov gamma telescopes. Using CORSIKA-simulated shower parameters, it finds that SVM with standardized data achieves the highest performance, and data transformations do not significantly affect model accuracy (p = 0.3165).

ABSTRACT

Cherenkov gamma telescope observes high energy gamma rays, taking advantage of the radiation emitted by charged particles produced inside the electromagnetic showers initiated by the gammas, and developing in the atmosphere. The detector records and allows for the reconstruction of the shower parameters. The reconstruction of the parameter values was achieved using a Monte Carlo simulation algorithm called CORSIKA. The present study developed multiple machine-learning-based classification models and evaluated their performance. Different data transformation and feature extraction techniques were applied to the dataset to assess the impact on two separate performance metrics. The results of the proposed application reveal that the different data transformations did not significantly impact (p = 0.3165) the performance of the models. A pairwise comparison indicates that the performance from each transformed data was not significantly different from the performance of the raw data. Additionally, the SVM algorithm produced the highest performance score on the standardized dataset. In conclusion, this study suggests that high-energy gamma particles can be predicted with sufficient accuracy using SVM on a standardized dataset than the other algorithms with the various data transformations.

Motivation & Objective

• To develop machine learning-based classification models for high-energy gamma particle detection using Cherenkov telescope data.
• To assess the impact of various data transformation and feature extraction techniques on model performance.
• To compare the performance of multiple machine learning algorithms on reconstructed shower parameters from CORSIKA simulations.
• To determine whether data preprocessing enhances classification accuracy for gamma particle detection.
• To identify the optimal model and data preprocessing strategy for high-precision gamma ray classification.

Proposed method

• Utilized Monte Carlo simulations via CORSIKA to reconstruct electromagnetic shower parameters from high-energy gamma particles.
• Applied multiple data transformation and feature extraction techniques to the simulated dataset to improve model generalization.
• Trained and evaluated diverse machine learning algorithms, including SVM, on both raw and transformed datasets.
• Standardized the dataset using z-score normalization to improve model convergence and performance.
• Employed paired statistical tests to compare model performance across different data preprocessing variants.
• Measured model performance using two distinct evaluation metrics to ensure robustness of results.

Experimental results

Research questions

• RQ1 Does applying data transformation techniques significantly improve the performance of machine learning models in classifying high-energy gamma particles?
• RQ2 Which machine learning algorithm achieves the highest classification accuracy on standardized and raw data?
• RQ3 How do different feature extraction methods affect the predictive performance of classification models?
• RQ4 Is there a statistically significant difference in model performance between raw data and transformed data?
• RQ5 Can SVM outperform other algorithms in detecting high-energy gamma particles when applied to standardized datasets?

Key findings

• Data transformations did not significantly affect model performance, with a p-value of 0.3165 from pairwise comparisons.
• SVM achieved the highest performance score on the standardized dataset compared to other algorithms and preprocessing variants.
• No significant difference was found between the performance of models trained on raw data versus transformed data.
• The study confirms that SVM with standardized data is the most effective configuration for high-energy gamma particle classification.
• The results suggest that extensive data preprocessing may not be necessary for optimal model performance in this context.
• The performance metrics indicate strong classification capability, with SVM demonstrating superior robustness across different data configurations.