The increasing need for effective caregiver training, particularly for Personal Support Workers, has led to the development of innovative e-training platforms. This study explores the application of advanced ML models to predict training outcomes and identify at-risk learners early in the process. The primary goal is to improve training completion rates while ensuring compliance with industry standards. We employed a range of ML models, including Decision Trees, Random Forest, Support Vector Machines, Neural Networks, to predict the likelihood of successful course completion using a dataset comprising over 27 million user interaction records. Feature engineering was used to extract key metrics such as module and lesson completion ratios. The results indicate that the Multilayer Perceptron model performed best, achieving an AUC score of 0.99, while K-NN also demonstrated strong performance with an AUC of 0.98. Key features such as module completion ratio and temporal progress were found to be significant predictors of training success. These findings suggest that integrating predictive analytics into e-training platforms can significantly enhance the effectiveness of PSW certification processes, ultimately supporting the growing demand for skilled caregivers in healthcare.