The machine-learning flare identification models for the article "Scalable, Advanced Machine Learning-based Approaches for Stellar Flare Identification: Application to TESS short-cadence Data and Analysis of a New Flare Catalogue"

This repository contains the machine learning methods for our multi-algorithm approach of flare identification in light curve data.

Models

We trained three models using three different algorithms:

1. Deep Neural Network (DNN)
2. Random Forest (RF)
3. XGBoost

These models are designed to identify flares in TESS short-cadence light curve data but can theoretically be applied to light curve data observed at different cadences. The models are user-friendly and can run on standard machines. You can find them in the "ML_models" directory of this repository:

DNNClassifier_flare_classification-by-Lin.keras

RandomForestClassifier_flare_classification-by-Lin.pkl

XGBoostClassifier_flare_classification-by-Lin.json

Tutorial

A comprehensive tutorial on how to effectively use these models is provided in "Tutorial.ipynb". This tutorial will guide you step-by-step on:

• Collecting flare candidates: How to gather flare candidates from the TESS light curve data.
• Feature extraction: How to determine the features of these flare candidates.
• Identifying true flares: How to use our machine learning models to identify "True Flares" among these candidates.

Installation and Dependencies

To run the models and tutorial, ensure you have all standard/wide-used Python packages (i.e., numpy, scipy, matplotlib, etc.) and the following dependencies installed:

• Python 3.x
• TensorFlow or PyTorch (for DNN)
• Scikit-learn (for Random Forest)
• XGBoost

You can install these dependencies using the following command:

Usage

To learn how to use the models, follow these steps:

1. Clone the repository:
   git clone https://github.com/CLL-Lin/MLsFlares.git

   cd MLsFlares
2. Follow the instructions in "Tutorial.ipynb" to start.

Citation

If you find our models useful, please cite the article "Scalable, Advanced Machine Learning-based Approaches for Stellar Flare Identification: Application to TESS short-cadence Data and Analysis of a New Flare Catalogue".

Note: This article is still under review. We will update the DOI and the announcement when the paper is published in Atronomical Journal.