Higher education institutions design faculty development programs consisting of a set of training courses. These programs are organized into training pathways that are offered during the academic year. The pathways allow participants to develop skills on various educational topics. Therefore, faculty members are expected to continually update their skills to improve their teaching practices. However, selecting a training course that is part of a pathway with multiple options is a difficult task for instructors because it is a time-consuming task. To provide instructors with training courses tailored to their interests and needs, we evaluated four collaborative filtering models based on matrix factorization using implicit feedback on a dataset of 2,113 instructors course enrollments from a faculty development program. The results are encouraging; a Weighted Approximate-Rank Pairwise loss model achieves 0.86 area under the curve score and a top-3 precision of 0.24, while an Alternating Least Squares model achieves 4.62 novelty metric. We discuss how the two models with the best performance operate in a real-world situation using their metrics results in order to understand the recommendations offered by each model to the user.