Boosting Model of Bioinspired Algorithms for Solving the Classification and Clustering Problems

Abstract

The paper considers the methods of boosting models application for solving clustering and classification problems. The main characteristics of boosting models are analyzed in the paper. The authors present the classification and clustering problems statement, describe popular modern and classical algorithms and analyze their benefits and shortcomings. According to the conducted research, a modified boosting model to solve the classification and clustering problems is developed and presented in the paper. The authors also compare the approaches of boosting and bagging and demonstrate their strengths and weaknesses. The paper describes the algorithms to be used in the developed boosting model. A new model of solving optimization problems is based on the usage of a weighted set of bioinspired clustering algorithms and their boosting. The heuristic of the suggested boosting method involves the use of a probability matrix providing a weighted estimation of the results obtained by different learning algorithms to achieve the highest quality of the problem solution. The developed approach is based on the usage of weighted data sets containing the probability of adding each individual element in a particular cluster. The conducted experimental research has shown that the developed boosting approach allows us to obtain the solutions equal or superior to those obtained by the popular algorithms.