Very cool. If you increased the noise variable upon initially creating your data, I wonder if the radial basis model would still fit the best. I'm curious to understand more how these models would perform under less cyclical timeseries data.