An Approach for Exoplanet Detection using Machine Learning By IJISRT

Abstract

Traditionally, astronomer and astrophysicist teams were the only ones capable of identifying planets. They used techniques and equipment that are only available to persons with years of formal education and training. NASA's program for exploring exoplanets has provided cutting-edge satellites that can gather a wide range of information on celestial objects to help with studies on these items. The capacity to write and understand machine-learning models has made planet detection more accessible to those using satellite data. This study used a number of classification methods and datasets to determine the likelihood that an observation is an exoplanet. The best machine learning model for classifying items of interest in the Cumulative Kepler Object of Information table was determined to be a Random Forest Classifier. The cross-validated accuracy of the Random Forest classifier yielded a score of 98%. The likelihood of 968 candidate observations being exoplanets is greater than 95%. Ultimately, an Azure Container Instance web service and an application programming interface (API) on the Microsoft Azure cloud enabled the Random Forest classifier to be publicly accessible.