Exploring Customer Behavior Through Push Notifications: A Comparative Analysis of Machine Learning and Deep Learning Approaches

In the context of the digitization of the retail market, mobile applications represent a strategic channel to improve customer interaction and optimize marketing strategies. This thesis project, carried out in collaboration with a large retail company, aims to analyze push notifications sent via the company's app and their impact on customer purchasing behavior, with the aim of optimizing their effectiveness. The work consists of three main objectives. First, an exploratory analysis of the open rates of push notifications was performed, divided into specific groups on the basis of subject categories and operational characteristics, to identify distinctive behavioral patterns in different customer segments. Next, the relationship between notification open rates and customer spending behavior was investigated to understand how engagement with notifications influences purchasing habits and the value generated. Finally, predictive models were developed to estimate open rates in different customer segments, with the aim of optimizing the sending of notifications and improving the targeting of communications. For the predictive task, machine learning models, such as logistic regression and XGBoost, and deep learning models, such as TabNet, were implemented. These models exploited the tabular format of the data to extract relationships between variables. The results obtained provide significant insights on the dynamics of the interaction between customers and push notifications. In particular, the integrated approach adopted makes it possible not only to customize notifications according to customer behavior, but also to increase the return on investment of promotional campaigns and improve loyalty.