In today’s world, diverse customer data, including demographic and behavioral information, enable companies to implement targeted marketing strategies. However, time and financial constraints make it impractical to reach all customers, and not all customers contribute equally to business value. This study proposes a profit-oriented ensemble approach for improving customer response modeling and prediction, where the “profit” variable, defined according to each customer’s individual value, is directly integrated into model training. The ensemble employs a neural network as the Meta learner, with backpropagation applied for weight updates. To address class imbalance in the target variable, observation weighting is incorporated. The model was evaluated on two public banking datasets and compared with traditional methods. Results indicate that the average profit from correctly predicted positive responses increased from 19.25 to 24.26 for the first dataset, and from 588.18 to 713.27 for the second dataset. These findings underscore the potential of profit-oriented ensemble modeling to enhance the effectiveness of marketing campaigns.

Keywords