Pattern Classification Program

The Pattern Classification Program is an open-source software intended for supervised and unsupervised pattern classification using vector measurements. It offers a range of algorithms and methods such as Fisher's linear discriminant, k-means clustering, and dimension reduction with Singular Value Decomposition. It also features a neural network in the form of Multi-Layer Perceptron, least-squares linear discriminant, Support Vector Machine algorithm, and more. Additionally, the software includes cross-validation, feature subset selection, parametric classifiers, and Bayes error estimation.

The program has two modes of processing, interactive and batch processing. Users can issue commands through a keyboard-driven menu system in the interactive mode while executing commands through batch files in batch mode. The software is straightforward and does not need any special run-time environment. PCP accepts tab-delimited text files for input data and saves results both on-screen and on a text file.

The Pattern Classification Program works on Linux and Windows operating systems with the i386 architecture CPU such as Intel Pentium or AMD Athlon. Developers tested and developed the software using RedHat Linux 9.0 distribution. It is also compatible with SUSE Linux 9.1, Fedora Core 2, Knoppix 3.7, and Windows XP.

The latest release comes with several improvements and bug fixes. For example, model selection support is available for the linear SVM kernel, and an option to build SVD transforms using a training or test dataset is also available. Additionally, P-errors are now reported in SVM model selection, and the build process was simplified.