Predicting the Next Transaction on Anonymized Payment Datasets with Deep Learning Models

Predicting customer behavior has long been a critical area of exploration for many companies, who often analyze purchase history to uncover behavioral trends and enhance their services. However, analyzing large amounts of personal customer data while maintaining compliance with data protection regulations (GDPR or LGPD) is challenging. In this paper, we propose three models that tackle the complexities of recognizing purchasing patterns for diverse applications in anonymized data. First, we evaluate architectures leveraging DL models for predicting subsequent purchase transactions using a dataset that safeguards confidential customer data while adhering to data protection regulations. The suggested models rely on Recurrent Neural Networks (RNN), Long Short-Term Memory (LSTM), and Gated Recurrent Units (GRU) to discern behaviors within a dataset devoid of personal information, allowing for comparison with other models pursuing the same goal. Then, we optimize each model's parameters, with findings indicating that the GRU-based model demonstrates superior generalization capabilities.