Exoplanets are among the most studied and remarkable topics in astronomy. Over the years, various methods have emerged for exoplanet detection, allowing for the identification of numerous exoplanet types. In this context, remote sensing and machine learning, which are central to our research, have significantly accelerated the detection process by leveraging algorithms. Our study involved training several machine learning models, including XGBoost, Random Forest, Multilayer Perceptron, K-Nearest Neighbor, Logistic Regression, and Support Vector Classifier, to compare their performance in both habitability assessment and exoplanet detection. The research utilized machine learning models trained on space observation data obtained from NASA, with the Python programming language serving as the foundation for the system's infrastructure. Our hypothesis was that "The detection of exoplanets and their evaluation within the scope of the habitability criterion can be increased to high accuracy rates with machine learning." Unlike merely detecting exoplanets, this study specifically aimed to identify Earth-like exoplanets. The XGBoost algorithm emerged as the most successful model in determining habitability, achieving an accuracy rate of 97.46% and demonstrating high precision and sensitivity. For exoplanet detection, all models achieved a main test accuracy rate of 96%; however, when considering sensitivity and precision, XGBoost was again the most effective. This research, following the synthesis and analysis of these two parameters, achieved a very high success rate compared to previous studies and made a significant contribution to the astronomy/astrophysics literature. Additionally, a Graphical User Interface (GUI) was developed, making the tested models functional through an application. The study successfully reached its goal of contributing important findings to the field.
Read full abstract