Definition of TDEs

The universe is constantly in motion – even stellar orbits in the center of galaxies are dynamic. The tidal disruption of stars occurs at the center of galaxies in the presence of supermassive black holes (SMBH). Some stars may be perturbed into almost radial orbits and pass within RT∼R*(MBH/M*)1/3 of the galactic central black hole (Gezari 2014). Here the black hole’s tidal force exceeds the star’s self-gravity, causing the star to undergo spaghettification as it is torn apart by the tidal forces. Half of the stellar mass is ejected from this system during disruption, while the remaining mass stays bound on initially-Keplerian orbits (Holoien et al. 2014). A transient accretion disk forms as this bound stellar debris fall back into the SMBH. This accretion disk then powers a luminous energetic flare. The fallback rate is predicted to follow the power-law of t−5/3, though deviations may occur. This sequence of events, from the disruption of the star to the fallback mass and luminous flare, is a tidal disruption event (Wu et al. 2018).

TDE Significance

SMBHs can be found at the center of every galaxy. However, they remain elusive because they are unobservable with light unable to escape its event horizon. The only way to observe an unobservable object is through its interactions with other observable objects. The detection and analysis of TDEs is a useful avenue for studying the properties of SMBHs, because the black hole spin and mass can be determined by the light emitted during the TDE flare (Mockler et al. 2019).

TDE Distinctions and Classification

TDEs seem to most often occur in quiescent Balmer-strong galaxy hosts, specifically in post-starburst galaxies. A sample of quiescent Balmer-strong galaxies in the Sloan Digital Sky Survey (SDSS) contained more than one-third of observed TDE host galaxies (Law-Smith et al. 2017). This suggests a dramatically increased rate of occurrences of TDEs in such galaxies. Post-starburst and other quiescent Balmer-strong galaxies are expected to host TDEs 20-200 times more than other galaxies (French et al. 2018).

TDEs are expected to observationally produce distinct light-curve evolution (photometric data) as seen in Figure 1 as well as spectral emissions of abrupt flux increase (spectroscopic data). These characteristics may then be used to distinguish them from both supernovae and normal active galactic nuclei. TDEs are further distinguishable by relative location to center of host galaxy and roughly t−5/3 fallback rate for lumosity decay. Despite these distinctions, TDEs and supernovae are commonly mistaken for each other (Law-Smith et al. 2017).

Random Forest Classifier

The Random Forest classifier generates a large number of decision trees for the given feature. Each tree “collaborates” with the others within the forest to determine the final classification. A subsample is selected from the training sample in the generation of each tree. Random cuts from the available features are tested. In this way, this method prevents overfitting because each tree evaluates a subsample of the features and data (French et al. 2018).

Overfitting occurs when a machine learning model is over trained on the training data, resulting in a model successful in classifying only the training data, but not at classifying new data. The Random Forest algorithm can classify objects using a large number of features. The high accuracy algorithm is not at risk of overfitting the classifier to the specific training sample. For these reasons, the Random Forest method is preferentially used in astronomical classification of luminous objects, including variable stars and quasar candidates (Dubath et al. 2011). It is, therefore, an ideal model to use for the classification of TDEs.

Black holes present the opportunity to understand the curvature of spacetime. Black holes are difficult to observe due to the inherent strong gravitational effects which not even light can escape. However, observing how blackholes interact with visible matter may provide insight into the detection and property dynamics of massive black holes. Tidal disruption events (TDEs) occur when a star’s path encroaches on the event horizon of a supermassive black hole, such that tidal forces pull the star apart in a process known as spaghettification. As a result, TDEs produce energetic jets that are hypothesized locations of high-energy neutrino production. This project presents a new strategy for efficiently classifying TDEs by employing dual machine learning Random Forest classifiers. Trained on ~11,000 photometric data points and ~20,000 spectroscopic data points, the classifiers analyzed photometric and spectral data independently. The results of the two classifiers were then used to classify the probability that an event is a TDE. The result of this project is a machine learning classifier with a mean accuracy of ~86.5475%. This new classification system will aid in more efficient and accurate classification of TDEs, allowing for greater understanding of supermassive black holes and the creation of galaxies.

There are several considerations to be further investigated. A mean value approach was used to replace missing data points, which theoretically should not affect the data significantly as light curves are continuous. To produce more robust results, a KNN machine learning algorithm or a Random Forest model could be utilized to fill a missing value. While these methods may be computationally expensive, this may be feasible due to the given small dataset. Furthermore, another method may be devised to utilize the outputs of the spectroscopic classifier and the photometric classifier as inputs into another machine learning program to create an even more robust model.