This paper presents a new method for solving a classification problem; the  BireyselValue method assumes that the individual traits of a class help to classify an observation based on similarity measures. The method involves three stages to solve the classification problem:  the building stage, the training stage, and the prediction stage. The first two stages accomplish two key steps: firstly, five parameters are used to transform any observation of size \(n\) variables into six variables; secondly, subsets of the individual traits of each class are created. As a result, the parameters, the subsets of the individual traits, and a scaled version of the training dataset are saved as a predictive model. Ultimately, the prediction stage uses the elements in the predictive model to transform the observations that are to be classified and of size \(n\) into the size of six variables and to perform similarity measures between the observation and the individual traits of class to make the final prediction. The experimental results obtained on 6 multiclass datasets from different domains showed that the proposed method is efficient at solving classification problems. Moreover, the method can potentially be used for purposes other than solving a classification problem.

Keywords: BireyselValue Method, Classification, Prediction, Dimension Reduction