An Automatic Classification Algorithm for Light Curves of Eclipsing Binary Stars Based on SVM

This paper proposes an automatic classification algorithm for light curves of eclipsing binary stars based on machine learning. At first the algorithm normalizes the light curves and performs filtering/interpolation to reduce the noise effect in the preprocessing stage, then the Fourier coefficients, which are extracted by FFT from the light curves, are used as the feature vector to train SVM and a classification model is obtained. We implement this algorithm with Python and use the data captured from CALEB/GCVS to validate and discuss the effect of the featured vector, the SVM kernel function and the penalty coefficient on the classifying accuracy. The classifying accuracies for the model on the training set and test set are 92.8% and 89%, respectively. Finally, we use a third-party data to verify the classification model and get a classifying accuracy of 88.8%. The results prove the validity of the classification algorithm proposed in this paper.