Machine learning in galaxy groups detection

Ibrahem, Rafee Tariq (2017). Machine learning in galaxy groups detection. University of Birmingham. Ph.D.

Preview

Ibrahem17PhD.pdf
PDF - Accepted Version
Download (2MB)

Abstract

The detection of galaxy groups and clusters is of great importance in the field of astrophysics. In particular astrophysicists are interested in the evolution and formation of these systems, as well as the interactions that occur within galaxy groups and clusters. In this thesis, we developed a probabilistic model capable of detecting galaxy groups and clusters based on the Hough transform. We called this approach probabilistic Hough transform based on adaptive local kernel (PHTALK). PHTALK was tested on a 3D realistic galaxy and mass assembly (GAMA) mock data catalogue (at close redshift z < 0:1) (mock data: contains information related to galaxies' position, redshift and other properties). We compared the performance of our PHTALK method with the performance of two versions of the standard friends-of-friends (FoF) method. As a performance measures, we used the precision versus recall curve. Furthermore, to test the efficiency of recovering the galaxy groups' and clusters' properties, we also used completeness and reliability, fragmentation and merging, velocity and mass estimation of the detected groups. The new PHTALK method outperformed the FoF methods in terms of reducing the detection of spurious agglomerations (false positives (FPs)). This smaller sensitivity to the false positive (FP) is mainly due to the clear description of the galaxy groups' model based on astrophysical prior knowledge; in particular, the fingers of god (FoG) pattern (a pattern formed by the projected velocity dispersion of galaxies, inside a galaxy group, along the line of sight). However, the FoF methods seem to outperform the PHTALK in terms of detecting galaxy groups or clusters that do not follow the FoG pattern. The main advantage of our probabilistic model is its flexibility to incorporate any prior knowledge expressed in terms of a galaxy group model.

Type of Work:

Thesis (Doctorates > Ph.D.)

Award Type:

Doctorates > Ph.D.

Supervisor(s):