Why not split your data into a training and testing set?
Then use a machine learning algorithm such as neural nets to assign random weights followed by a gradient descent.
Use test set and see if you like the model.
Unless you're mining for relationships that you don't know exist, in which...