I have decided to do undergraduate thesis on "estimating photometric redshift" or something related to this using machine learning. Reading previous papers, I have come to know that work has been done on this topic using support vector machine, artificial neural network, Bayesian approach and even deep learning. Can you suggest what are the modifications or additions that I can possibly make?
The process of determining photometric redshift includes, e.g.,: 1. objects of interest 2. types of data (e.g., one epoch multiple filters, one filter multiple epochs) 3. method of analysis (including model, and algorithm)
From what you mentioned, those are just algorithms. So, you might take a look around on other aspects as listed.