Neural networks based on the supervised and unsupervised learning methodologies have been tested for their classification and clustering abilities in the normal as well as hostile conditions. A problem from the NMR (nuclear magnetic spectroscopy) was used as the test for this and it was found that the neural network based on both the paradigms fared better withou feature enhancement. The neural network tuning is the most difficult part in the connectionist pos approach and some sugggestions from this side is incorporated to obtain optimum results.

For more details, refer this link:
http://pubs.acsdoi/pdf/10.1021/ci00021a019?cookieSet=1
also refer this pdf:
http://citeseerx.ist.psu.edu/viewdoc/dow...1&type=pdf