Basis Function Construction in Reinforcement Learning Using Cascade-Correlation Learning Architecture