Machine Learning for Exoplanet Detection: A Comparative Analysis Using Kepler Data

TL;DR

This study evaluates various machine learning algorithms for exoplanet detection using Kepler data, finding Random Forest to be the most accurate and robust method, especially when combined with SMOTE for class imbalance.

Contribution

It provides a comprehensive comparison of ML classifiers for exoplanet detection, highlighting the effectiveness of ensemble methods like Random Forest with data balancing techniques.

Findings

Random Forest achieves 99.8% accuracy.

SMOTE improves model performance significantly.

Ensemble methods outperform simpler classifiers.

Abstract

The discovery of exoplanets has expanded our understanding of planetary systems and opened new avenues for astronomical research. In this study, we present a machine learning (ML) framework for exoplanet identification using a time-series photometric dataset from the Kepler Space Telescope, comprising 3,198 flux measurements across 5,074 stars. We investigate the performance of four supervised classification algorithms, namely Random Forest, k-Nearest Neighbors (KNN), Decision Tree, and Logistic Regression, using a comprehensive set of evaluation metrics such as accuracy, precision, recall, F1-score, Area Under the Receiver Operating Characteristic Curve (AUC-ROC), confusion matrices, and learning curves. Among the models, Random Forest achieves the highest accuracy (99.8\%) and near-perfect F1-scores, demonstrating superior generalization and robustness. KNN also performs strongly, achieving 99.3\% accuracy, while Decision Tree demonstrates moderate performance with 97.1\% accuracy, and Logistic Regression trails behind with the lowest accuracy and generalization at 95.8\%. Notably, the application of the Synthetic Minority Over-sampling Technique (SMOTE) significantly improves performance across all models by addressing class imbalance. These findings underscore the effectiveness of ensemble-based machine learning techniques, particularly Random Forest, in handling large volumes of photometric data for automated exoplanet detection. This approach holds significant potential for implementation at ground-based facilities, such as the Iranian National Observatory (INO), where such extensive and precise datasets can further advance exoplanet discovery and characterization efforts.