# Classification of a black hole spin out of its shadow using support vector machines

## TL;DR

This study employs Support Vector Machines to classify black hole spins from simulated shadow images, demonstrating high accuracy especially with disk matter models and highlighting the influence of matter distribution, viewing angle, and resolution.

## Contribution

The paper introduces a machine learning approach using SVMs to classify black hole spin from simulated images, analyzing the impact of matter distribution and image resolution.

## Key findings

- High classification accuracy (>90%) for disk matter models at various resolutions.
- Classification accuracy drops below 80% for spherical matter models at low resolutions.
- Image resolution, matter distribution, and viewing angle significantly affect spin classification accuracy.

## Abstract

We use Support Vector Machines (SVMs) to classify the spin of a black hole. The SVMs are trained and tested with a catalog of numerically generated images of black holes, assuming disk and spherical matter models with monochromatic emission with wavelength of 4mm. We determine the accuracy of the SVM to classify the spin in terms of the image resolution, for which we consider three resolutions of $16^2,~32^2$ and $64^2$ pixels. Our approach is applied to the specific mass of the Supermassive Black Hole (SMBH) at the center of the Milky Way. Our findings are that when the distribution is a thin disk, the accuracy in the classification resists even the coarsest resolution with accuracy over 90\%, whereas for the spherical distribution it drops below 80\% for low and intermediate resolutions. The results show how the distribution of matter, the angle of vision and the image resolution influence the ease to determine the black hole spin.