Grupiranje srednjoškolaca i studenata prema njihovim interesima, fobijama i navikama klasterskom analizom

Klasterska analiza metoda je grupiranja objekata u klase tako da su u istoj klasi objekti najveće sličnosti. U ovom radu, klasterska analiza primijenjena je na podatcima o interesima, hobijima i zdravstvenim navikama studenata i srednjoškolaca. Podatci su prikupljeni koristeći online anketu koja je distribuirana putem društvenih mreža. Konačan broj analiziranih ispitanika je bio 699, pri čemu je omjer studenata i srednjoškolaca ravnomjeran. Za primjenu metode, potrebno je izabrati odgovarajuću mjeru sličnosti i algoritam ovisno o vrsti objekata, svojstvima algoritama te konačnim ciljevima istraživanja. Klasterski algoritmi mogu biti hijerarhijski ili nehijerarhijski (partitivni). Hijerarhijski algoritam je primijenjen koristeći Ward metodu, a rezultati su prikazani dendrogramom. Kao partitivni algoritam izabran je k-means s konačnim ograničenjima na dva, a potom i na tri klastera, dok je mjera sličnosti euklidska.Cluster analysis is a method of grouping objects into classes so that the objects with the highest similarity are in the same class. In this paper, cluster analysis has been applied on a data set which consists of interests, hobbies and health habits of students and high school pupils. The data set was collected using an online survey which was distributed through social media. The final number of analyzed respondents was 699, having the ratio between students and high school pupils relatively equal. In order to apply the method, it is necessary to choose an adequate similarity measure and an algorithm depending on the type of data, features of the algorithm and main goals of the research. Clustering algorithms are either hierarchical or non-hierarchical (partitive). A hierarchical algorithm has been applied using the Ward method, and the results were shown on dendrograms. As for the type of partitive algorithm, the k-means algorithm was chosen, limiting the final number of clusters on two and on three, with the euclidean similarity measure