Kifle, Mesfin (PhD)Abebaw, Banchalem2021-11-292023-11-292021-11-292023-11-292021-04-07http://etd.aau.edu.et/handle/123456789/29026Almost in every discipline, people are using phones that generate detail records containing information’s about phone usage, such as, connected time of the call, the identities of sources, the identities of destinations, the duration of each call, the amount billed etc. named as (customer detail records) CDR. It’s difficult for someone to scan through all the data and establish the relative decision for the business because it is time. As a result, there is a growing interest towards better solutions for finding, organizing and analyzing these CDR data in telecommunication companies. The effective ways of organizing telecom data form later decision making and business management efficient, less complicated, friendly and low-cost. Customer detail record (CDR) clustering is one of the common methods of managing customers in business using their usage behavior in network. This study proposes a model that cluster telecommunication customers using detail records (CDR) of the user. During the clustering process, all the CDR records pass through pre-processing stages to prepare data for processing. Then transformation of preprocessed data was done by through scaling algorithm. The scaled features were extracted from the sampled CDR record. Finally, customers are clustered based on the usage behavior using the most popular K-means algorithm that is based on the cosine similarity of the weighted features. Ethio-telecom customer detail records were used for experimentations. The clustering results are evaluated to find optimal cluster size using elbow and silhouette methods. The result shows the value of silhouette coefficient is greater when cluster size is four, with this we clustered the sampled data into four classes. When we see the number of customer’s distribution in each class from hundred thousand sampled data, in the first class 19.959 % of the sample was grouped. In the second class 0.603% of the sample was grouped together. In the 3rd class there are 4.383% of the sample with related usage characteristics. In the 4th class there are 75.054% customers with related usage characteristics. The four groups are considered as customers with low usage history of the service, common usage history of the service, customers with very good usage history and customers with very midst usage history. The ultimate goal of any business would be to have as many customers up there in the service category. This can be further used in decision making to have many numbers of customers to our needed group i.e., a group with very good usage behavior of services.enCDR ClusteringCustomer Data AnalysisFeature ScalingK-means ClusteringThesis