friday7/exoplanet-detection

Exoplanets are defined as planets orbiting a star outside our solar system. There have been considerable improvements over the past decades in the detection of exoplanets. NASA launched the Kepler satellite in 2009 with the mission to hunt exoplanets. The data gathered by the satellite have been made publicly available by NASA in an attempt to increase development in the field of astroinformatics, a crossdisciplinary field consisting of astronomy, data science and informatics. This has led to increased contribution in detecting exoplanets among non-exoplanets. In this paper, we explore various techniques to detect exoplanets using the intensity of light captured by Kepler space telescope. The dimming in flux (light intensity) of the stars indicates an orbiting body around that star. We perform preliminary analysis and prepare the data to be fed in to our data mining models. We have used multiple models to efficiently classify the exoplanets. The first model, H2O’s Deep Learning, is based on a multi-layer feedforward artificial neural network. We have then performed the Gradient Boosting technique using XGBoost algorithm, an ensemble technique that works on the concept of Decision Tree and Bootstrap Aggregation. Finally, we have used Convolutional Neural Network, a Black Box method, to classify the exoplanets. This paper will provide the results, model performance and efficiency of the various algorithms used to detect such exoplanets.