I'm currently working on a project that's aiming to compare broker results using both classification and regression algorithms. I'm not a TAC expert so I'm here to ask for help and clarification.
1. Is it a good approach to use classification/regression algorithms to train and predict tariff values and prices? I'm afraid that learning from past executions may lead to bad decisions or my broker being too slow.
2. If its good to apply those algorithms what is the best set of variables to be used/considered when developing my learning base? Is it possible to get them from state and trace log files?
3. There is any recommended algorithm to be used? Any material (papers and so on) covering past execution learning on TAC problems?