Improving Telecommunications Upselling through Explainability and Model Simplification

11/27/2024

In the telecommunications industry, upselling is crucial for increasing revenue and offering higher-value offers to customers. Using machine learning in this context allows for more accurate predictions of which offers are most likely to attract specific customers. To improve the accuracy of these predictions, complex models are often used, which can lead to a lack of transparency in decision-making, creating a barrier to their acceptance and application. Therefore, it is essential to develop solutions that not only produce accurate results but also provide relevant explanations. This thesis, based on data from a Portuguese telecommunications company, evaluated the explainability of the model currently used for upselling and proposed adjustments to enhance its interpretability. The methodology involved two approaches: one focused on explainability, using SHAP-based methods, and another aimed at improving the model by reducing its complexity and the number of variables. In terms of explainability, notable adjustments were made to an explainable clustering approach, making it suitable for large datasets. The findings showed that reducing the model’s complexity and the number of variables improved both performance and explainability. Additionally, the proposed methodology offered clearer and more accessible explanations than methods that rely solely on SHAP values. As a result, business teams can now more easily understand how the model operates and make more informed decisions regarding the offers that are presented to customers.

Authors in the community:

• 

  J

  José Maria Nogueira de Pinho Marques

  ist195751

Supervisors of this institution:

• 

  M

  Maria da Conceição Esperança Amado

  ist13493

mathematics - Mathematics

por - Portuguese

Instituto Superior Técnico