The MSE fitness function provides a root mean square (of the error) output for a given training set. The value of NATURAL.ORDER.STRATEGY
will tell the genetic algorithm to handle fitnesses slightly differently. If it is true, it will make fitnesses that approach 0 better (as in the case of the XOR example)
or if false, the the larger the value returned the better the fitness, which in the context of an error minimising function is not really relevant and is included for completeness.
All you need to do to use this in your experiment is to set the relevant data source in the TRAINING.SET field of .net file, create the other config files (you can use the XOR examples as a template) and that's it.
You can try and predict stocks, commodities, weather etc. Obviously, the accuracy of the output very much depends on the quality of the input data and the difficulty of the problem.
Note, the only data files currently supported are csv files. However, there is nothing to stop you wrinting your own data loader, see Creating Data Loaders for more details.