E-commerce has emerged as a powerful platform for revenue generation in business. Various sales strategies have been adopted to maximize revenue while meeting customer expectations. Recent advancements in machine learning and deep learning have played a crucial role in enhancing e-commerce sales. This research explores the potential of employing machine learning techniques to provide personalized product bundle recommendations based on customer preferences, aiming to improve cross-selling and gross merchandise volume. However, machine learning models often lack explainability, which leads them to be termed black box models. Currently, there is a lack of explainable recommendation models that offer insights into the inner workings of deep learning models. To explore the product relationships and to understand and support the decision-making processes for different recommendation models, a visual analytics system is built. This thesis explains and interprets deep learning-based recommendations by visualizing the relationships and interactions within customer-product data. The visualization tool is built upon pilot interviews conducted from a business perspective to provide insights to stakeholders and enhance the interpretability of recommender systems. We explore the use of machine learning models for automatic product bundling, benchmark the results, and use visualization with the help of dimensionality reduction for high dimensional data, hierarchical edge bundling, scatter plots, collapsible trees, and heatmaps. The implementation involves employing JavaScript libraries, Python, HTML, and CSS. The visualization toolbox is built upon the requirements gained through a pilot interview. It was evaluated via an expert study that analyzed the visualization toolbox with parameters like transferability and evocativeness, reusability, abundance, and understandability.