An automated knowledge extraction system
Youngjun Lee · 1995
With the advent of computer technology, many organizations have more data than they can deal with. The number and size of available databases is growing so fast that there will never be enough human analysts to interpret all the data. Therefore, what we need is intelligent systems that analyze data to discover patterns, regularities, and knowledge. The cost of computing is less than that of a professional's time spent on analyzing data and the cost of collecting data. The system thus has to discover only a few unexpected relationships to pay for itself. Extracting knowledge from the data can be automated, making a significant impact on the human decision-making process. We developed a Fuzzy Generalized Exemplar (FGE) learning and inference model. It produces generalized exemplars. Each is augmented with its own distance metric. Each generalized exemplar can be considered a fuzzy rule. Human subjects are highly flexible in that they can selectively weight an attribute differently depending on the region of the instance space in which it is located. Like humans, the FGE based learning system can allow the importance of attributes to be a function of its region of instance space. Based on the FGE model, we designed and implemented a system to extract fuzzy rules from data. These fuzzy rules capture the essential information contained within the patterns and relationships in the data. The rules can be used in two ways. The first way is to use the rules directly to interpret and understand the active mechanism underlying the data. The second way is to apply the rules to new data and predict new outcomes. We performed extensive comparisons of the system with several advanced machine-learning approaches and human experts, where such comparisons were possible. The simulation results demonstrate that an FGE based learning model can learn effectively in a diverse set of domains. The performance of the system compares favorably with that of other machine learning methods and human experts.