Robert, one way to fix things is to use more data. For example you could have used half of your dataset to find your pattern/correlation, and then used the other half to test it. Having derived the hypothesis from the full dataset, the only thing to do is to get more data and test it on that instead.