| Description |
Detecting the background galaxies within the spectrum of the foreground galaxy is one of the most effective ways to identify strong lensing phenomena. However, it is very hard and time consuming for astronomers to apply this search method manually (i.e., one by one) to huge cosmological datasets. This study attempts to predict the background galaxies and discover the potential lensed candidates by using classification methods. To achieve this, the most important step is to leverage cosmological data by extracting potentially useful features for the classification methods. In this study, after extracting the potentially useful features from two different astronomy datasets, chi square weighting feature selection was applied to them to find the final set of the useful features. Then, various state-of-the-art classification methods were applied on the datasets to predict lens candidates. Classifier performance was measured in terms of accuracy, Area Under the Curve (AUC), and F-measure. The results showed that 85 features chosen by chi square weighting are the most useful features. Logistic Regression outperformed all other classification methods for the prediction task. Finally, the prediction method using classifiers is significantly more efficient than manual inspection. The proposed method in this study is generalizable for detecting background galaxy and potential lenses in any cosmological data. This can significantly improve the efficiency for astronomers to apply their search methods. |
