While this sounds interesting, the feature selection was done on the same data as the tests were conducted later on.

If I read it correctly, that's a biggie, especially given the weak outperformance of chance.