The University of Wisconsin-Madison is considered the number one public research university in the nation. A Chronicle of Higher Education report ranked UW-Madison as the fourth largest recipient of overall research and development funds. Founded in 1848, it is home to more than 43,000 students, and hosts the fifth largest population of international students in the country. The UW Medical School and School of Nursing have worldwide reputations for extensive contributions to the advancement of medicine and health. With more than $64 million in federal awards each year, it is one of the leading biomedical research institutions in the country. The Research Design and Statistics Unit (RDSU) is a technical support unit for both the schools of medicine and nursing providing high levels of statistical and design support for ongoing and proposed research.
Numerous researchers were searching for a tool to assist them in developing insight into the problems under study, or to “learn from their data.” Classical biostatistical modeling provided too many limitations for researchers in the early stages of their investigation. While neural networking tools were abundant and extensively used, most researchers were uncomfortable with the lack of detail provided by such an approach. The PolyAnalyst suite enabled our researchers to continue to empirically search their data for rules and structure while providing a symbolic knowledge of the structure, the detail they needed. Currently PolyAnalyst suite is being used with some success on research in Nephrology, Pediatrics and Phonology and Communicative Disorders. PolyAnalyst has assisted in obtaining various mathematical representations of problems in these projects and has provided guidance for researchers in furthering their investigations. For example, PolyAnalyst has allowed modeling kidney survival time with a high degree of explanation. These results have been presented at a major international medical conference.
The PolyAnalyst suite offers a range of capabilities, including data access, data manipulation and cleaning, machine learning, visualization, and reporting. PolyAnalyst can directly access data held in Oracle, DB2, Informix, Sybase, MS SQL Server, Ingres, or any other ODBC-compliant database. Data and exploration results can also be exchanged with MS Excel 7.0 or 97. PolyAnalyst also provides for data subsetting and variable transformations. PolyAnalyst’s major feature is its broad selection of self-learning engines for data analysis, which include; PolyNet Predictor, a hybrid of GMDH (Group Method Data Handling) and neural networking approaches, Find Laws, which is based on evolutionary programming technology, Find Dependencies exploration engine, Classify engine, Discriminate engine, Cluster engine, and Multiparametric Linear Regression which discovers linear relations in data. PolyAnalyst also provides an interactive graphics representation of data and findings and allows report generation, all in a point-and-click environment.
PolyAnalyst’s major strength lies in the Find Law engine utilizing Symbolic Knowledge Acquisition Technology (SKAT), which automatically discovers dependencies and rules in data and represents these rules in explicit mathematical statements. The ability to present a simple mathematical relation discovered in the data with an assessment of the appropriateness of this mathematical function is extremely valuable to researchers. Also the suite provides more strengths by including other exploration engines. Another major function provided by PolyAnalyst is the ability to graphically interact with the data and results of the discovered functional relationships. This makes for a complete exploratory assessment of data.
One weakness in the tested version (PolyAnalyst Pro Version 3.5), was the lack of the ability to cut and paste graphical representations from PolyAnalyst to other software.
The selection criteria used by the RDSU hinged primarily on the location of rules and the presentation of the findings in symbolic mathematical expressions, along with an interactive graphical interface.
The goal of the RDSU is to provide the most up-to-date data analytic support to its researchers, PolyAnalyst has provided another dimension for that support.
Megaputer Intelligence support has been timely, with a knowledgeable staff more than willing to assist in the use of the product.
Documentation is good, and along with Megaputer Intelligence’s web site, examples, datasets, and technical information are very useful and easily available.
Dr. R. L. Brown
Professor and Director
Research Design & Statistics Unit
University of Wisconsin-Madison
School of Nursing and Medicine