A differential evolution algorithm for data clustering

U radu je opisan problem grupiranja podataka, popularni algoritam za grupiranje k-means te algoritam diferencijalne evolucije (DE), popularan evolucijski algoritam koji može biti primijenjen za grupiranje podataka. Implementirane su dvije inačice algoritma DE koje se razlikuju u korištenom operatoru mutacije, te je dan opis tog programskog rješenja, način rada i prikaz njegove uporabe. S implementiranim programskim rješenjem provedena je eksperimentalna analiza na nekoliko sintetičkih skupova podataka i na nekoliko često korištenih stvarnih skupova podataka. U analizi su uspoređeni algoritam k-means i dvije inačice algoritma DE. Vrednovane su njihove učinkovitosti za grupiranje podataka u smislu mjera SSE i DBI. Rezultati prikazuju da u većini slučajeva inačice algoritma DE pronalaze particije s boljim vrijednostima DBI, dok algoritam k-means pronalazi particije s boljim vrijednostima SSE. Razlike između inačica algoritma DE za korištene skupove podataka su zanemarive.The paper describes the concept of data clustering, popular algorithm for clustering k-means and differential evolution (DE) algorithm which is a popular evolution algorithm that can be used for data clustering. There are two versions of DE algorithm implemented which differ in the mutation operator. A description is given of the implemented software solution, the way it works and how it is used. Using the implemented software solution, an experimental analysis was carried out on several synthetic datasets and on several often used real world datasets. This analysis compared the quality of partitions acquired by the k-means algorithm and the two versions of DE algorithm. The measures used for the quality of partitions were sum of squared errors (SSE) and DaviesBouldin index (DBI). The analysis showed that both versions of DE algorithm got better DBI results, while k-means algorithm got better SSE results. The differences in results between the two versions of DE algorithm for the datasets used were negligible