Abstract:
Businesses today need to know their customers to maintain a competitive margin and run effectively. Customer Segmentation helps companies to understand their customer base such as their behaviors, demographics, and psychographics etc. In addition, the MultiOffer Recommendation model predicts purchasing patterns for non-bundled customers, if these customers will subscribe to one or more bundles. To understand customer segments, comparative analysis was conducted between DBSCAN and K-means clustering. While DBSCAN excelled in identifying noise points, its clustering performance was ineffective as it failed to capture discrete customer segments. On the contrary, K-means demonstrated satisfactory performance by clearly identifying heterogeneous clusters, facilitating insightful post-cluster analysis which helps understand different type of customers. Manual campaigns were conducted post cluster analysis and found to be effective overall in comparison with current manual campaigns. For the Multi-Offer Recommendation model, Random Forest Classifier and Random Forest Regressor with Multi-Output Classifier and Multi-Output Regressor were used. Post development marketing campaigns found that regressor based campaigns were more successful than classifier-based campaigns in terms of response rate, indicating their suitability for this dataset.
