Improving the reliability of photometric redshift with machine learning

In order to answer the open questions of modern cosmology and galaxy evolution theory, robust algorithms for calculating photometric redshifts (photo-z) very large samples galaxies are needed. Correct estimation various photo-z algorithms' performance requires attention both metrics data used estimation. this work, we use supervised machine learning algorithm MLPQNA calculate in COSMOS2015 catalogue unsupervised Self-Organizing Maps (SOM) determine reliability resulting estimates. We find that spec-z<1.2, predictions on same level quality as SED fitting photo-z. show SOM successfully detects unreliable spec-z cause biases performance. Additionally, select objects with reliable predictions. Our cleaning procedures allow extract subset which final catalogs is improved by a factor two, compared overall statistics.