|
|
| Identifier |
000425744 |
| Title |
Spectral classification of stars based on Machine Learning methods |
| Alternative Title |
Φασματοσκοπική ταξινόμηση αστεριών βασιζόμενη σε μεθόδους μηχανικής μάθησης |
|
Author
|
Κυρίτσης, Ηλίας Μ.
|
|
Thesis advisor
|
Ζέζας, Ανδρέας
|
|
Reviewer
|
Παυλίδου, Βασιλική
Reig, Pablo
|
| Abstract |
High Mass X-Ray Binaries (HMXBs) are systems that consist of a compact object (Neutron Star or Black Hole) and a massive companion star with O- B-spectral type.The knowledge of the spectral types of these stars is crucial because it can provide us a wealth of information about the formation and evolution of HMXBs systems. Previous years, big surveys were dedicated in the spectral classification of the companions star in these systems either in the Galaxy or in the Magellanic Clouds. Spectral classi¬fication was performed through visual examination of their spectra ,according to the presence or the absence of characteristic spectral lines. Nowadays, where the number of spectroscopic data is continuously this approach is time consuming and suffers from subjectivity. Thus, the need of an new objective automated method is more timely than ever. In this work, we use the popular supervised machine learning algorithm Random Forest to develop an automated spectral classifier for early type stars. In our sample are included 777 OB stars from different surveys. We measure the Equivalent Width of 18 characteristic spectral lines (features) following a scheme developed for the classification of these sources. We optimized our model by searching for the best values of the hyperparameters as well as the best combination of spectral lines. We reached a prediction accuracy ~ 70 % using 14 out of the initial 18 lines in our scheme. Finally , we apply our model in a sample of 28 sources both from the Galaxy and the Small Magellanic Cloud, with known spectral types from visual inspection, reaching a success rate of ~ 60%.
|
| Language |
English |
| Subject |
HMXBs |
|
High energy astrophysics |
|
Machine learning |
|
Random forests |
|
Αστροφυσική υψηλών ενεργειών |
|
Διπλά συστήματα εκπομπής ακτίνων Χ |
|
Τυχαία δάση |
| Issue date |
2020-03-27 |
| Collection
|
School/Department--School of Sciences and Engineering--Department of Physics--Post-graduate theses
|
|
|
Type of Work--Post-graduate theses
|
| Views |
832 |
Spectral classification of stars based on Machine Learning methods
