Publicação
Feature selection strategies for improving data-driven decision support in bank telemarketing
| Resumo: | The usage of data mining techniques to unveil previously undiscovered knowledge has been applied in past years to a wide number of domains, including banking and marketing. Raw data is the basic ingredient for successfully detecting interesting patterns. A key aspect of raw data manipulation is feature engineering and it is related with the correct characterization or selection of relevant features (or variables) that conceal relations with the target goal. This study is particularly focused on feature engineering, aiming at the unfolding features that best characterize the problem of selling long-term bank deposits through telemarketing campaigns. For the experimental setup, a case-study from a Portuguese bank, ranging the 2008-2013 year period and encompassing the recent global financial crisis, was addressed. To assess the relevance of such problem, a novel literature analysis using text mining and the latent Dirichlet allocation algorithm was conducted, confirming the existence of a research gap for bank telemarketing. Starting from a dataset containing typical telemarketing contacts and client information, research followed three different and complementary strategies: first, by enriching the dataset with social and economic context features; then, by including customer lifetime value related features; finally, by applying a divide and conquer strategy for splitting the problem in smaller fractions, leading to optimized sub-problems. Each of the three approaches improved previous results in terms of model metrics related to prediction performance. The relevance of the proposed features was evaluated, confirming the obtained models as credible and valuable for telemarketing campaign managers. |
|---|---|
| Autores principais: | Moro, Sérgio Miguel Carneiro |
| Assunto: | Telemarketing Banking Bank deposits Decision support Decision making Feature selection Feature engineering Data mining Banco Depósitos bancários Sistemas de suporte à decisão Apoio à decisão Seleção de atributos Engenharia de atributos |
| Ano: | 2015 |
| País: | Portugal |
| Tipo de documento: | tese de doutoramento |
| Tipo de acesso: | acesso aberto |
| Instituição associada: | ISCTE |
| Idioma: | inglês |
| Origem: | Repositório ISCTE |
| Resumo: | The usage of data mining techniques to unveil previously undiscovered knowledge has been applied in past years to a wide number of domains, including banking and marketing. Raw data is the basic ingredient for successfully detecting interesting patterns. A key aspect of raw data manipulation is feature engineering and it is related with the correct characterization or selection of relevant features (or variables) that conceal relations with the target goal. This study is particularly focused on feature engineering, aiming at the unfolding features that best characterize the problem of selling long-term bank deposits through telemarketing campaigns. For the experimental setup, a case-study from a Portuguese bank, ranging the 2008-2013 year period and encompassing the recent global financial crisis, was addressed. To assess the relevance of such problem, a novel literature analysis using text mining and the latent Dirichlet allocation algorithm was conducted, confirming the existence of a research gap for bank telemarketing. Starting from a dataset containing typical telemarketing contacts and client information, research followed three different and complementary strategies: first, by enriching the dataset with social and economic context features; then, by including customer lifetime value related features; finally, by applying a divide and conquer strategy for splitting the problem in smaller fractions, leading to optimized sub-problems. Each of the three approaches improved previous results in terms of model metrics related to prediction performance. The relevance of the proposed features was evaluated, confirming the obtained models as credible and valuable for telemarketing campaign managers. |
|---|