Enhancing Photometric Redshift Predictions and Uncertainty Quantification Using Deep Learning Methods

In the realm of astronomical data, data-driven techniques promise to predict photometric redshifts more efficiently than traditional methods such as template fitting. We propose a neural network model with probabilistic layers, leveraging data from the Sloan Digital Sky Survey. Two model variants are explored: one with uncertainty predictions and another without. Our findings highlight insights from uncertainty estimates, elucidating confidence in redshift predictions and facilitating informed decision-making. The model without uncertainty predictions achieves an RMSE value of approximately 0.23, showcasing that data-driven approaches have the efficiency and adaptability to advance redshift prediction in the rapidly growing field of astronomical data.