Analysis of Product Recommendation Models at Each Fixed Broadband Sales Location Using K-Means, DBSCAN, Hierarchical Clustering, SVM, RF, and ANN

Abstract

The telecommunications industry proliferates in the digitalization era, especially Fixed Broadband services. Fast and stable internet access is essential, especially at sales locations with appropriate products. This research aims to develop an optimal product recommendation model for each sales location, using machine learning with a mixed method approach, with a combination method of clustering and classification, where the clustering method is used for the geographic segmentation stage. Then, the results of each cluster from the geographic segmentation are used as input for the classification method, which is a stage called sales forecasting. Next, the performance analysis measured the accuracy level of each combination of models. The best model combines clustering and classification models, which, on average, across all clusters, gives the best accuracy value. The data used in this research is GIS-based POI data and sales history data, which is internal data from a telecommunications company in Indonesia. From the tests carried out in this research, the best model combination is the K-Means and the Random Forest models, with an accuracy value of 82.08%. Meanwhile, the lowest performance resulted from a combination of the K-Means and ANN models with an accuracy value of 79.50%. With an average combination model performance above 80%, this research shows that using mixed methods with clustering and classification can provide valuable insights in subsequent research, especially in the context of the telecommunications industry, especially in fixed broadband services.