1. The regression for the redshifts of galaxies in SDSS DR18
- Author
-
Xiao-Qing, Wen, Hong-Wei, Yin, Feng-Hua, Liu, Shang-Tao, Yang, Yi-Rong, Zhu, Jin-Meng, Yang, Zi-Jie, Su, and Bing, Guan
- Abstract
•There was rarely such massive data to validate the efficiency of algorithms. The Sample 1 and 2 were from the SDSS DR18 combined with the ALLWISE. They were very large and new.•The LightGBM, XGBoost, Catboost, RF, DF, DT, KNN, GBDT and SVR algorithms were used to regress the redshifts of galaxies. These were very popular machine learning methods.•In the LightGBM, we got MAE ∼ 0.018, MSE ∼ 0.001, root mean squared error (RMSE) ∼ 0.033, MAPE ∼ 0.146, R2 ∼ 0.909, bias∼ 0.000233, σ∼ 0.033389, σMAD∼ 0.0181, outliers fraction ∼ 0.0028, the ratio of |∆z| > 2σ∼ 0.0245, biasnorm∼ -0.000404, RMSEnorm∼ 0.025, σNMAD∼ 0.015774 and run time ∼ 31.021s for our Sample 1. Our results were a little better than other's results.•We got MAE ∼ 0.044, MSE ∼ 0.007, RMSE ∼ 0.083, MAPE ∼ 0.149, R2 ∼ 0.895, bias ∼ 0.000489, σbias∼ 0.083065, σMAD∼ 0.0352, outliers fraction ∼ 0.0186, the ratio of |∆z| > 2σ∼ 0.0396, biasnorm∼ -0.00222, RMSEnorm∼ 0.053, σNMAD∼ 0.026093 and run time ∼ 61.847s for our Sample 2. The redshifts of galaxies in the main part with higher redshifts in the Sample 2 were also regressed well up to z <1.
- Published
- 2024
- Full Text
- View/download PDF