DIFFUSE NEAR K-NEIGHBORS METHOD FOR CLASSIFYING HOMOGENIZED INFORMATION BY A DISTANCE-BASED ON PREFERENCE AND RATIOS.

摘要

The present invention relates to a method of mining data for scanning information in a massive repository and providing elements for making decisions in any application domain. The said method finds knowledge in the form of classification models that predict a value, called a class, for an attribute, belonging to a record to be classified which demands an estimated value starting from the historical records, called predictor attributes. In addition, it solves four problems: 1) heterogeneity of the predictor attributes; 2) skewed estimation of numerical predictor attributes; 3) ignoring the relevance of predictor attributes; 4) conflict when choosing the class. Respectively solves 1) transforming categorical information to numerical values with range [0, 1] using fuzzy logic criteria; 2) to standardize the numerical information through ratios of the range [0, 1]; 3) distinguishing the semantic and context relevance of the attributes according to the criteria of the user who provides qu alitative values that are converted to numerical weights ranging [0.5 to 1.5] to expand or reduce the numerical distance separating the values of a predictor attribute; 4) choosing the class, subset of neighboring K neighboring registers, the distance of its predictor attributes with respect to the ones of the record to be classified is the