| Abstract: | Customer segmentation helps organizations group similar customers, aiding in tailored marketing strategies. Mobile money services, like CBEBirr by the Commercial Bank of Ethiopia, are widely used in Ethiopia, with 10.2 million customers utilizing CBEBirr for services such as cash in, cash out, send money, buy airtime, pay bills, buy goods, and other financial services. Previously, CBEBirr customers were not segmented to get information. Thus, our study explores using unsupervised machine learning to segment CBEBirr customers at the Commercial Bank of Ethiopia. In this study, CBEBirr customers are segmented according to their similarities based on demographics, including age, gender, and data. CBEBirr customers are recruited by agents, branches, and merchants. Besides demographic data, behavioural data such as the number of cash-in transactions, cash-out transactions, send money transactions, buy air time transactions, pay bill transactions, and buy goods transactions of the customer were also used. The segmentation model is done using four unsupervised machine learning algorithms: K-means clustering, agglomerative clustering, density-based spatial clustering of applications with noise (DBSCAN), and mean shift, using 170,012 CBEBirr customers' data gathered from Commercial Bank of Ethiopia. To evaluate the performance of the developed model, the two most popular evaluation metrics for clustering algorithms, the silhouette coefficient, and the Davies-Bouldin index, were used. We obtained silhouette scores of 0.792, 0.676, -0.129 and 0.792 and Davies Bouldin Scores of 0.291,0.290,1.306, and 0.290 for K-means clustering, agglomerative clustering, DBSCAN, and the mean shift algorithm, respectively. Hence, this concludes that considering those evaluation metrics among the four algorithms we used to cluster our CBEBirr customer’s data, the mean shift algorithm is better than agglomerative clustering, DBSCAN, and the K-means algorithm, as displayed by the high value of the silhouette score and the low value of the Davies Bouldin score. The study will support Commercial Bank of Ethiopia in gaining knowledge about its CBEBirr customers, specifically which services of CBEBirr are more commonly used by the customers, and the bank will formulate marketing strategies accordingly, in turn achieving its goal of making a cashless society. |
