<p>Understanding customer purchase behavior and accurately predicting it is essential for enhancing marketing effectiveness and ensuring the success of e-commerce platforms. However, two major challenges must be addressed for these solutions to be effective: the feature engineering of complex customer behavior data and the inherent class imbalance present in transactional data. This Systematic Literature Review (SLR) examines 47 studies published from 2019 to early 2025. The review synthesizes existing approaches, identifies knowledge gaps, and outlines future research directions concerning feature engineering and class imbalance in the context of e-commerce purchases. Key findings reveal substantial progress in feature engineering techniques, including deep learning methods, the incorporation of contextual and loyalty features, graph-based feature extraction, and advanced selection methods. While strategies such as SMOTE and ensemble learning are widely utilized to tackle class imbalance, there is an increasing interest in combined sampling strategies, despite their validation still being somewhat limited. In prediction models, Ensemble Learning techniques, such as XGBoost and Random Forest, along with Deep Learning approaches, are widely used and often outperform traditional methods when optimized effectively. However, there is no single best model; the ideal choice is contingent upon the application context, data characteristics, and the effectiveness of feature engineering and imbalance management. This systematic literature review highlights the interactions among feature engineering, imbalance handling, and model selection, and identifies key research gaps, including the need for improved validation of imbalance techniques and the integration of demographic data. Future research should focus on more comprehensive strategies to improve purchase prediction in dynamic e-commerce settings.</p>