• Data Mining
• Forgotten Items
• Machine Learning
• Next-Basket Prediction

Next-Basket Prediction (NBP), which provides customers with recommendations for future purchases based on their shopping habits, has been an important area of research in retail analytics. As the field evolves, two significant challenges have emerged: the identification and prevention of forgotten items - products that customers intended to purchase but overlooked during their shopping trip - and the growing complexity of prediction models that makes their decision-making processes increasingly opaque. This thesis addresses these challenges by first establishing a formal definition and evaluation framework for the task Forgotten-Item Prediction (FIP). We then propose novel methodologies that not only outperform baseline methods in identifying forgotten items but also incorporate interpretability mechanisms. These methods generate data-driven explanations for their predictions, providing customers with clear insights into why specific items are predicted as likely to be forgotten. Through extensive experimentation on real-world retail datasets, we demonstrate both the superior predictive performance of our proposed approaches and their ability to provide meaningful, actionable explanations for their predictions.