engleski

Comparison of Clustering Algorithms for Optimal Restaurant Location Selection Using Location-Based Social Networks Data

Machine learning algorithms are increasingly used in various fields. Unlike supervised algorithms that require the engagement and knowledge of experts in a particular area, unsupervised algorithms do not need it and are therefore more comfortable to use. Clustering algorithms belong to unsupervised algorithms and are used to group data according to a given similarity criterion with achieving significant similarity between data within the same group and minor similarities between data belonging to different groups. In this paper, five clustering algorithms in restaurant location optimization in Zagreb are analyzed. The clustering algorithms' output result lists municipalities in Zagreb city divided into groups with similar properties. Based on these data, the investor can quickly conclude what individual municipalities are similar and based on that, a more objective assessment of the location of a restaurant or catering facility can be made before the investment. The data based on which the algorithms divided parts of Zagreb into groups were obtained from a social network that can store user locations. One of the essential functions of the used social network is sharing information about restaurants, cafes, and other catering facilities. The common name of these social networks is a location-based social network. The paper compares the Gaussian Mixture Model algorithm, k- means algorithm, Hierarchies algorithm, Agglomerative Clustering algorithm, and Spectral Clustering algorithm. The selected five algorithms have the property that one of their input variables is the number of clusters.

Clustering ; Big data ; Restaurant ; Foursquare ; Location-based social network

nije evidentirano