Data Access
Feature transformation
Feature selection
Learning algorithms
Error evaluation


SIPINA uses an owner file format (.FDM) which is optimized for I/O operation. SIPINA can also import and export the datasets int other file format such as : text file, C4.5 file format, etc. If you import a new file, the first column can be considered as label of examples.

You can edit the dataset into a grid : add/remove examples, add/remove attributes, modify values, etc.

If you are running an analysis, you cannot modify the current dataset.


SIPINA has a set of attributes transformation methods. You can : discretize an attribute, coding a set of attributes from a discrete attribute, etc.


SIPINA can automatically select the most informative attributes for a supervised analysis. There is "filter methods" which select the best predictive attributes independently of supervised algorithms used, before the induction (e.g RELIEF, MIFS, etc.). There is also "wrapper methods" which choose the best attributes using the supervised algorithm, all algorithms available into SIPINA can be used in the wrapper session.


SIPINA proposes various supervised learning algorithms, a detailled list is available in "Learning algorithms" options of the main menu of this site (see SIPINA LEARNING ALGORITHMS).

Note that the choosen algorithm determines the kind of predictive attributes you can introduce in your analysis (e.g. Neural Network accepts only continuous predictive attributes).

Each induction method can be embedded in a meta-learning session, such as aggregation methods (bagging, arcing, boosting) or filtering non-informative examples (John's Robust algorithm), etc.


SIPINA can estimate error rate on a test set, i.e. you have carried out a partition of the dataset "learning set / test set" before the induction process. Several resampling error estimate (cross-validation, bootstrap, etc.) are also available.


SIPINA can classify examples of a new unclassified dataset with learned classifiers. Name of attributes must match between learning set and the dataset you want to classifiy.