Neural networks are alternative predictive techniques that have significant advantages and disadvantages as compared with traditional statistical techniques. Neural networks or neural classifier systems attempt to identify patterns and cluster observations together in large data sets. They can take many independent variable inputs and predict a dependent output. The most significant advantages of neural networks over traditional statistical techniques is that they can recognize patterns that are extremely complex and they are not negatively affected by collinearity of variables. Neural networks do have drawbacks, as they can be extremely computationally intensive to train, and they only work with large, clean data sets.
While once a largely academic technique, virtually every credit card transaction in the US is now run through a neural network that examines patterns of usage on the card for expected behavior that could indicate fraud.