Previous Page | Noesis Main Page | Next Page

Noesis Supervised Pattern Recognition (SPR)




Supervised Pattern Recognition (SPR) is the process by which a data set is partitioned into various groups (clusters) using the properties of an "example" partitioning in some other data. This assumes that some data with the a desired clustering (partitioning) has previously been achieved (manually or via Unsupervised Pattern Recognition methods). The SPR algorithm is then trained, based on this known data set and partitioning. Once trained, the algorithm can partition new data to clusters similar to those of the example classification. All necessary preprocessing of the new data is also done automatically. Trained algorithms can be saved to (or retrieved from) files. The user has control over several parameters and can investigate the effectiveness of the training process (how well the algorithm is trained). All actions are easily undone to provide a high level of flexibility and user friendliness. Several SPR algorithms (including neural networks) are included. Following is a list of some of the basic functions and features available with Noesis' SPR:

  • SPR Wizard allowing the user to easily "cruise" through the various choices to train an SPR method (classifier). The wizard provides all relevant information (method, parameters etc) so that SPR training and results overview becomes simple.

  • Automatic Usage Set (unknown data) pre-processing based on Noesis Script Log (feature selection, normalization, feature extraction etc.).

  • Multiple SPR algorithms including Neural Networks (k-NNC, BP Net etc.).

  • Interactive SPR algorithm training and testing modes.

  • Multi-mode independent Training and Testing data sets.

  • Classification result output to PAC (DTA, TDA or WFS) files (see also Data Handling).

  • Descriptive statistics for clustering evaluation (see also Statistics).