Recommender System for mobile subscriber provisioning.
| dc.contributor.author | Sibanda, Elias Mbongeni | |
| dc.contributor.co-supervisor | Thomas, Varghese | |
| dc.contributor.supervisor | Zuva, Tranos, Prof | |
| dc.date.accessioned | 2022-01-27T23:56:23Z | |
| dc.date.available | 2022-01-27T23:56:23Z | |
| dc.date.issued | 2018-04 | |
| dc.description | M. Tech. (Department of Information and Communication Technology, Faculty of Applied and Computer Sciences) Vaal University of Technology. | en_US |
| dc.description.abstract | Mobile phone recommendation systems are of great importance for mobile operators to achieve a profit. In a user-derived market, the number of contract users and contract phones is especially significant for mobile service operators. The tremendous growth in the number of available mobile cellular telephone contracts necessitates the need for a recommender system to help users discover suitable contracts on the basis of their usage patterns. Recommender systems recommend items to users and their primary purpose is to increase sales and recommend items that are predicted to be suitable for individual users. There are two commonly used techniques in developing recommender systems including collaborative- and content-based filtering. Recommender systems make their recommendations based on data that is available on the system. These systems have gained popularity over the years and they have been adopted in many domains. In this study a recommender system for mobile subscriber provisioning was developed using a hybrid J48 and kmeans algorithms. The J48 algorithm was used for classifying subscribers per usage stream and then k-means was used to cluster all the subscribers of similar usage patterns. The algorithms were selected after being compared with other algorithms and the two performed best in their categories. The clustering algorithm, k-means, was able to cluster the sample data as follows: Cluster 0 contained 48% (1621) of the subscribers cluster 1 contained 42% (1423) subscribers, cluster 2 contained 8% (272) subscribers and lastly cluster 3 contained 74 subscribers representing 2% of the population and the run time of k-means is faster than that of EM. The classification algorithm j48 performed at an average of 99.98% for correctly classifying instances and this was higher than the Naïve Bayes, zeroR and MLP algorithms. The developed recommender system was able to successfully recommend contract packages to subscribers. A precision-recall curve was produced, and it showed good performance of the system. This study successfully highlighted the challenges in recommender systems, and showed that a hybrid system was better able to recommend products to the mobile subscribers. | en_US |
| dc.identifier.uri | http://hdl.handle.net/10352/478 | |
| dc.language.iso | en | en_US |
| dc.subject | Mobile phones | en_US |
| dc.subject | Algorithm | en_US |
| dc.subject | Recommender system | en_US |
| dc.subject.lcsh | Dissertations, Academic. | en_US |
| dc.subject.lcsh | Mobile computing. | en_US |
| dc.subject.lcsh | Mobile communication systems. | en_US |
| dc.title | Recommender System for mobile subscriber provisioning. | en_US |
| dc.type | Thesis | en_US |