Habitability Classification of Exoplanets: A Machine Learning Insight

Abstract

We explore the efficacy of machine learning (ML) in characterizing exoplanets into different classes. The source of the data used in this work is University of Puerto Rico's Planetary Habitability Laboratory's Exoplanets Catalog (PHL-EC). We perform a detailed analysis of the structure of the data and propose methods that can be used to effectively categorize new exoplanet samples. Our contributions are two fold. We elaborate on the results obtained by using ML algorithms by stating the accuracy of each method used and propose the best paradigm to automate the task of exoplanet classification. The exploration led to the development of new methods fundamental and relevant to the context of the problem and beyond. Data exploration and experimentation methods also result in the development of a general data methodology and a set of best practices which can be used for exploratory data analysis experiments.

Evolution of Novel Activation Functions in Neural Network Training with Applications to Classification of Exoplanets

“…Explicit Earth-similarity score computation (Bora et al, 2016) based on parameters mass, radius, surface temperature and escape velocity developed into Cobb-Douglas Habitability Score helped identify candidates with similar scores to Earth. However, using Earth-similarity alone (Agrawal et al, 2018) to address habitability is not sufficient unless model based evaluations (Saha et al, 2018c) are interpreted and equated with feature based classification (Basak et al, 2018). However, when we tested methods in (Basak et al, 2018), it was found that the methods didn't work well with pruned feature sets (features which clearly mark different habitability classes were removed).…”

“…However, using Earth-similarity alone (Agrawal et al, 2018) to address habitability is not sufficient unless model based evaluations (Saha et al, 2018c) are interpreted and equated with feature based classification (Basak et al, 2018). However, when we tested methods in (Basak et al, 2018), it was found that the methods didn't work well with pruned feature sets (features which clearly mark different habitability classes were removed). Therefore, new machine learning methods to classify exoplanets are a necessity.…”

“…features such as surface temperature and features related to surface temperature) are removed from the feature set. To this end, a version of ANNs, replicated weight neural network (RWNN), in conjunction with sigmoid activation, has been applied to the task of classification of exoplanets (Basak et al, 2018). However, the results turned out to be less than satisfactory.…”

“…Even though our focus is not all on the performance of statistical machine learning or other learning algorithms on the PHL-EC data, we feel it's necessary to report the performance of some of those briefly. This was a postfacto analysis (the realization dawned upon us much later) done to ascertain the efficacy of our methods compared to methods such as Gaussian Naive Bayes (GNB), Linear Discriminant Analysis (LDA), SVM, RBF-SVM, KNN, DT, RF and GBDT (Basak et al, 2018). These methods didn't precede the exploration into activation functions.…”

“…) in order to "learn" to perform classification/regression tasks by considering examples and without the aid of task-specific programming rules. As of today, ANNs are found to yield state-of-the-art performance in a variety of tasks such as speech recognition, computer vision, board games and medical diagnosis (Xiao and Lu, 2015;Narayanan et al, 2004) and classification of exoplanets (Basak et al, 2018).…”