Click Prediction in Digital Advertisements: A Fuzzy Approach to Model Selection

Abstract

The fuzzy logic theorem is inherently used effectively in expressing current life problems. So, using fuzzy logic in machine learning is getting popular. In machine learning problems, especially using digital advertisement data, products/objects are being trained and predicted together, but this can cause worse prediction performance. A significant commitment of our research is, we propose a new approach for ensembling prediction with fuzzy clustering in this study. This approach aims to solve this problem. It also enables flexible clustering for the objects which have more than one cluster’s characteristics. On the other hand, our approach allows us ensembling boosting algorithms which are different types of ensembling and very popular in machine learning because of their successful performance in the literature. For testing our approach, we used an online travel agency’s digital advertisements data for predicting each hotel’s next day click amount, which is crucial for predicting marketing cost. The results show that ensembling the algorithms with a fuzzy approach has better performance result than applying algorithms individually.