Mechanizing Hypothesis Formation

Mechanizing hypothesis formation, also known as the GUHA method, is a method of exploratory data analysis. Its development started in the 1960s and was inspired by the question "Can computers formulate and verify scientific hypotheses?" A GUHA procedure is an algorithm trying to answer this question. Its input consists of analyzed data, of several parameters defining a large set of relevant patterns and of a true condition. The output consists of all prime patterns satisfying the given true condition. A pattern is prime if it does not logically follow from another shorter true pattern already involved in the output.

Several GUHA procedures dealing with patterns in a form of rules, couples of rules, action rules, histograms, couples of histograms, or patterns based on general contingency tables are involved in the LISp-Miner system developed at the University of Economics, Prague. This system was is used many practical tasks of data mining as well as in teaching and in research.

The start of a cooperation with the OptiSolutions company was a milestone in the development of the LISp-Miner system. A result of this co-operation is the system OptiMiner which consists of new implementations of the analytical GUHA procedures originally developed for the LISp-Miner system. It is crucial that the OptiMiner is an inherent part of a large LOTYLDA platform developed by the OptiSolutions. This platform serves as a background for many real data mining and business applications as well as big data analysis. Thus, also the OptiMiner is used in various commercial applications.

The book covers

• a brief overview of theoretical foundations of the GUHA method
• examples of applications of the GUHA procedures to solve real problems relevant to big data analysis, data mining, and business intelligence
• an overview of the recent research results and examples of their applications.