About NeuNet Pro
Table Of Contents

About NeuNet

License Agreement


How to Purchase

Version History

Future Plans

NeuNet Pro Future Plans

  • Let Us Know!
    Please let us know what features you would like to see in future versions of NeuNet Pro. Post your ideas on the NeuNet Pro Forum. We depend on feedback from our users to help chart the course of NeuNet Pro into the future.

  • Data Shuffling
    Several users have requested the ability to apply random "shuffles" to their data. We are currently working on this feature and it will be available in the near future. (We have built a freeware data shuffler available by email request.)

  • Biasing for Classification
    In some classification projects it might be desirable to bias the prediction toward a particular class. For example in medical diagnosis a false negative may be much worse than a false positive. A simple tweak to the SFAM algorithm would allow the user to request biasing toward certain classes.

  • Inductive Rule Extraction
    We are experimenting with the ID3 and C.45 algorithms developed by J. Ross Quinlan at the University of Sydney. The ability to extract a decision tree from a database appears to be an attractive way to obtain predictions and perform data mining. Text fields could be used as inputs. Inductive Rule Extraction could be included along-side the SFAM and Back Propagation modules in NeuNet Pro, or could be packaged into a separate program. This algorithm allows text fields to be used as inputs. We are very excited about the possible uses of this feature. (We built this as a separate program. Visit the VisiRex Website.

  • Multiple Layer Back Propagation
    We are looking into the possibility of including multiple layer back propagation nets as well as other learning algorithms. However, we believe the real strength of NeuNet Pro is its relative simplicity for the user. We will endeavor to maintain our ease-of-use and will try to avoid needless complications.

  • Missing Values
    Currently, missing values are handled by skipping any record that contains missing values. The prediction for these records is shown as "N/A". The user may decide to configure an additional project to provide a predicted estimate for these missing values. A suggested option is that missing values could simply use the field average as a quick estimate.

  • Verify Data Rows
    Currently, the neural net is optimized by verifing on the training set. A separate "verify" set would help to avoid over-training.

A Complete Neural Network Development System

CorMac Technologies Inc.
34 North Cumberland Street ~ Thunder Bay ON P7A 4L3 ~ Canada
E m a i l