Supervised Classification Problem: Searching for Maximum Patterns

A logical approach to the supervised classification problem is considered. A well-known classifier model is studied, in which the analysis of initial integer data is reduced to the search for certain fragments in the feature descriptions of precedents, called maximum patterns. The main disadvantage of this model is the high number of classification refusals. In addition, the search for maximum patterns is a computationally complex discrete problem. In the presented work, a modification of the model learning stage is proposed based on a new, more effective way of enumerating maximum patterns. It is experimentally established that due to special linear ordering of feature values a higher quality of classification can be achieved.