A Study of Light Intensity of Stars for Exoplanet Detection using Machine Learning

Exoplanets is defined as the planets outside the solar system. With a hope to find a habitable, earth-like planet, the detection of exoplanet is very crucial domain of research. Many researchers and space agencies are working on manifold variation of strategies to find novel exoplanets. In this paper, we explore the detection of exoplanets using machine learning classification. We explore time-series based light intensity data of different stars to understand a pattern that correlates with presence of exoplanets in that star's orbit. The data used in this study was collected under the NASA Kepler mission - Campaign 3. After initial analysis of the said data, we further implemented various classification algorithm to detect exoplanets. The K-nearest neighbor model trained on SMOTE re-balanced data gave an accuracy of 98.20% with a F1-score of 98%. Further, understanding the difficulty of collecting data in real time, we considered implementing the classification models on limited data by limiting the observations of light intensity for each star to multiple thresholds between 75% to 1%. The accuracy and F1-score for even the model with only 1% data for KNN model was found to be 90.1% and 90% respectively.