Clustering bank clients into groups is becoming increasingly popular, as banks must approach clients more individually to remain competitive. By clustering clients into groups, they can develop a different approach to each group of clients, increasing both client satisfaction and their own profit. The master's thesis offers an approach to finding the optimal method for clustering NLB bank's customers based on two machine learning algorithms: the k-means algorithm and the k-medoids algorithm.
The critical part is selecting and identifying the model and its parameters. The clustering of the bank's clients is carried out several times, for different sets of input variables and different settings of the algorithm parameters, and we look for the model in which the clustering is the most successful. The clustering obtained is evaluated using methods for assessing the quality of clustering into groups, where the Silhouette and Dunn indexes are used. Based on the values of both indices, we choose the optimal combination of algorithm and parameters for clustering. Finally, we explain the optimal clustering with the help of the distribution of the most important variables, which we obtain based on the random forest algorithm.
|
