# A machine learning approach for identification and classification of symbiotic stars using 2MASS and WISE

## TL;DR

This paper develops machine learning methods to identify and classify symbiotic stars using infrared data from 2MASS and WISE surveys, improving detection accuracy and discovering new candidates.

## Contribution

The study introduces a novel machine learning framework utilizing classification tree, LDA, and KNN to distinguish symbiotic stars from similar objects in infrared surveys.

## Key findings

- Identified key infrared colour indices for symbiotic star classification.
- Recovered 90% of known Galactic symbiotic stars in tests.
- Discovered 72 new symbiotic star candidates.

## Abstract

In this second paper in a series of papers based on the most-up-to-date catalogue of symbiotic stars (SySts), we present a new approach for identifying and distinguishing SySts from other Halpha emitters in photometric surveys using machine learning algorithms such as classification tree, linear discriminant analysis, and K-nearest neighbour. The motivation behind of this work is to seek for possible colour indices in the regime of near- and mid-infrared covered by the 2MASS and WISE surveys. A number of diagnostic colour-colour diagrams are generated for all the known Galactic SySts and several classes of stellar objects that mimic SySts such as planetary nebulae, post-AGB, Mira, single K and M giants, cataclysmic variables, Be, AeBe, YSO, weak and classical T Tauri stars, and Wolf-Rayet. The classification tree algorithm unveils that primarily J-H, W1-W4 and Ks-W3 and secondarily H-W2, W1-W2 and W3-W4 are ideal colour indices to identify SySts. Linear discriminant analysis method is also applied to determine the linear combination of 2MASS and AllWISE magnitudes that better distinguish SySts. The probability of a source being a SySt is determined using the K-nearest neighbour method on the LDA components. By applying our classification tree model to the list of candidate SySts (Paper I), the IPHAS list of candidate SySts, and the DR2 VPHAS+ catalogue, we find 125 (72 new candidates) sources that pass our criteria while we also recover 90 per cent of the known Galactic SySts.