Classification of Fanaroff-Riley Radio Galaxies using Conventional Machine Learning Techniques

Large scale radio astronomy projects such as the Square Kilometre Array (SKA) is expected to produce extremely large data sets (300 PB p/a), which makes purely manual classification no longer a viable option. An automatic method of radio galaxy classification is proposed in this paper that will examine the performance of conventional classification approaches to determine whether accurate classification can be done on subjective features. We define a subjective features as a characteristic of the data that is not dependent on an objective or self-contained measure. This method is meant to act as a tool for domain experts or as an auxiliary function for deep learning classifiers. The morphological features extracted are lobe size, hot spot/peak brightness, the number of lobes present and the Fanaroff- Riley (FR) ratio, with the latter two viewed as objective features and the rest subjective features. To determine the relevance of the morphological features as indicators of class differences, the features are then used in several machine learning classifiers, of which the study compares the classification accuracy and F1-score. Overall, the Random Forest classifier produces the highest accuracy (94.66%) and F1-score (0.94) using all the features in conjunction. The relevance of the Fanaroff- Riley ratio as a classification metric is verified. Further investigation in automatic extraction of the ratio is required.