Blood donation is critical for ensuring a stable and reliable supply of blood, yet blood donor retention remains a complex and persistent challenge. Previous attempts to develop predictive models for blood donor retention have often yielded relatively low accuracy and fail to address the class imbalance challenge that come with blood donation data, limiting their practical application in addressing this challenge. This study investigates the use of the Light Gradient Boosting Machine (LightGBM) as a high-performance gradient boosting framework for predicting blood donor retention. LightGBM employs a leaf-wise growth strategy, which significantly improves accuracy by minimizing loss at each iteration. It also supports histogram-based learning, reducing memory consumption and speeding up computation, making it suitable for the blood donation prediction. The study utilized data obtained from Kenya blood banks, consisting of 5000 records and nine features, to develop and evaluate the model. The LightGBM model achieved an accuracy of 98.3% and an F1 score of 97.8 which was higher as compared to the existing models. The results demonstrate that LightGBM is an effective and computationally efficient tool for predicting blood donor retention. Its ability to handle large, imbalanced datasets and complex patterns makes it well-suited for real-world applications in predictive analytics. This study provides blood agencies with a more reliable model for accurately predicting blood donor retention, reducing recruitment costs, and enabling targeted retention strategies to ensure a steady blood supply.