Classification of Top Call Reasons using Machine Learning in Call Center Service
| dc.contributor.advisor | Dereje, Hailemariam (PhD) | |
| dc.contributor.author | Atsede, Abebe | |
| dc.date.accessioned | 2022-02-07T11:05:59Z | |
| dc.date.accessioned | 2023-11-04T15:13:02Z | |
| dc.date.available | 2022-02-07T11:05:59Z | |
| dc.date.available | 2023-11-04T15:13:02Z | |
| dc.date.issued | 2021-12 | |
| dc.description.abstract | A call center (CC) connects customers to a service provider, such as telecom operators. The CC receives of customer requests, feedback, and complaints. These inputs from the customers provide an opportunity to comprehend the customer’s needs, problems the customers face when using services, and the performance of the service provider. Meeting customer expectations by responding to complaints increases customer satisfaction, which translates to revenue maximization. Hence, the CC is critical to the success of the service provider. Ethio-telecom, a telecom service provider in Ethiopia, operates a large CC that provides telecom-related services throughout the country. The CC accepts over two million calls per day via a service-free line. The center records and maintains a huge amount of customer-related data, which can further be analyzed using state-of-the-art machine learning algorithms for the purpose of proactively estimating call types and reasons. This thesis proposes to map features from customer profile information into top call reasons so as to better understand customer call requests and map future calls to specific top call reasons. Data was extracted from Ethio-telecom’s IP contact center, customer relational management, and customer billing system servers. To construct the classification models, J48, Random Forest (RF), and Naive Bayes (NB) algorithms are used. Accuracy, time to build a model, and model interpretation of each algorithm are used to compare their performance. Results show that RF and J48 algorithms outperform NB, with scores of 97.46% and 97.4%, respectively. The NB model is the least accurate, with an accuracy of 83.6%. However, the time spent building a model for NB is less compared to J48. During the model's interpretation, J48 algorithm is more interpretable than the NB and RF. J48 algorithm are best. | en_US |
| dc.identifier.uri | http://etd.aau.edu.et/handle/123456789/29945 | |
| dc.language.iso | en_US | en_US |
| dc.publisher | Addis Ababa University | en_US |
| dc.subject | Classification algorithm | en_US |
| dc.subject | Call Center | en_US |
| dc.subject | IP Contact Center | en_US |
| dc.subject | Customer Relational Management | en_US |
| dc.subject | Customer Billing System | en_US |
| dc.subject | J48 | en_US |
| dc.subject | Naive Bayes | en_US |
| dc.subject | Random Forest | en_US |
| dc.title | Classification of Top Call Reasons using Machine Learning in Call Center Service | en_US |
| dc.type | Thesis | en_US |