School of Physics - Theses

Permanent URI for this collection

http://hdl.handle.net/11343/310

Filters

2022 1
Reset filters

Settings
Statistics
Citations
Type: PhD thesis × Start date: 2020 × End date: 2022 × Author: Alonzi, Noura Mohammad H ×

Search Results

Now showing 1 - 1 of 1
  • Item
    Thumbnail Image
    Finding Quasars in the Southern Hemisphere Sky Using Random Forest Machine Learning
    Alonzi, Noura Mohammad H ( 2022-04)
    Quasars are the most luminous persistent sources in the universe. The main goal of this thesis is to search for quasars at different redshifts using an efficient machine learning algorithm: the Random Forest classifier. A technique was developed and tested on a small photometric sample, the Early Data Release (EDR) from the SkyMapper Southern Survey (SMSS). After classification by Random Forest, candidates were prioritised for confirmation using observations from the ANU 2.3m telescope. SkyMapper is the first digital optical survey in southern hemisphere and has been used to build the required training subsets and the dataset. SMSS has been matched with other available surveys in the southern hemisphere to provide a broader range of colours for selection algorithms. The predictions were greatly improved by combining photometric colours in the optical from SMSS with mid-infrared data from AllWISE. Random Forest Machine Learning techniques provided classifications with probabilities of up to 81%. The EDR pilot study, predicted 119 QSO-candidates. Of these 78 have been confirmed as quasars, either previously or by new observations, and the remainder still need to be observed. So far, only one galaxy and one star were found amongst the candidate list. In addition, the classifier has been trained and applied to a much larger dataset, the Third Data Release of SMSS. This provides a preliminary study of the techniques that will be required to study extremely large samples of quasars identified in the Legacy Survey of Space and Time which will commence in a few years time on the Vera C. Rubin Observatory. Other observations in the thesis explored different AGN types, such as narrow line objects, using the BPT diagnostic diagram to determine the source of excitation energy. Future machine learning algorithms may be able to determine finer AGN classifications, as the range and quality of the input non-spectroscopic datasets improves. The conclusion from these studies is that candidate quasars can be identified with high confidence using machine learning, if a sufficiently large spectroscopic test sample is available.