Photometric Supernova Classification With Machine Learning

Abstract

Automated photometric supernova classification has become an active area of research in recent years in light of current and upcoming imaging surveys such as the Dark Energy Survey (DES) and the Large Synoptic Survey Telescope, given that spectroscopic confirmation of type for all supernovae discovered will be impossible. Here, we develop a multi-faceted classification pipeline, combining existing and new approaches. Our pipeline consists of two stages: extracting descriptive features from the light curves and classification using a machine learning algorithm. Our feature extraction methods vary from model-dependent techniques, namely SALT2 fits, to more independent techniques fitting parametric models to curves, to a completely model-independent wavelet approach. We cover a range of representative machine learning algorithms, including naive Bayes, k -nearest neighbors, support vector machines, artificial neural networks and boosted decision trees (BDTs). We test the pipeline on simulated multi-band DES light curves from the Supernova Photometric Classification Challenge. Using the commonly used area under the curve (AUC) of the Receiver Operating Characteristic as a metric, we find that the SALT2 fits and the wavelet approach, with the BDTs algorithm, each achieves an AUC of 0.98, where 1 represents perfect classification. We find that a representative training set is essential for good classification, whatever the feature set or algorithm, with implications for spectroscopic follow-up. Importantly, we find that by using either the SALT2 or the wavelet feature sets with a BDT algorithm, accurate classification is possible purely from light curve data, without the need for any redshift information.

Deep Recurrent Neural Networks for Supernovae Classification

“…For a smaller sample training set of 5.2% of all the data we again perform similarly to Karpenka et al (2013), but underperform compared to Newling et al (2011), taking into account the slightly larger sample size in the latter case. In Lochner et al (2016), using the SALT2 fits provided the best average AUC over a range of machinelearning techniques. By imposing a purity of 90%, a completeness of 85% was achieved, while requiring a completeness of 90% reveals a corresponding purity of 85%.…”

“…The analyses by Karpenka et al (2013) and Newling et al (2011) are easier to compare. Along with Lochner et al (2016), these employ a two-step process, where features are first extracted by various methods before machinelearning classification. The results obtained for similar sized training sets are comparable, as can be seen in the top section of Table 2.…”

“…Machine-learning methodology has previously been applied to SPCC classification (Newling et al 2011;Karpenka et al 2013;Lochner et al 2016). Instead of performing feature extraction before classification, our approach uses the light curves directly as inputs to a recurrent neural network, which is able to learn information from the sequence of observations.…”

“…However, it is estimated that only 5000 to 10,000 will be spectroscopically 1 confirmed by follow-up surveys (Matheson et al 2013), so classification methods need to be developed for photometry. All previous approaches to automated classification (Newling et al 2011;Karpenka et al 2013;Lochner et al 2016) have first extracted features from supernovae light curves before using machine-learning algorithms. One of the advantages of deep learning is replacing this feature extraction.…”

Measuring the Properties of Dark Energy with Photometrically Classified Pan-STARRS Supernovae. I. Systematic Uncertainty from Core-collapse Supernova Contamination

“…Machine learning techniques can often outperform these classifiers, yielding higher SN Ia classification efficiency (the fraction of SNe Ia classified correctly) and lower CC SN contamination (Lochner et al 2016;Möller et al 2016). On SDSS-SN data, the Sako et al (2014) kd-tree nearest neighbor (NN) method has a purity comparable to that of Campbell et al (2013) but accurately classifies ∼1.4 times as many real SNe Ia in a given sample.…”