Recommendation systems in libraries are increasingly being discussed in various studies related to improving library services. With the presence of Digital Libraries with a very large number of collections, it makes difficult for users to find the relevant collection. This condition provides an opportunity for libraries to provide more personalized services to their users by utilizing recommendation systems. The use of machine learning models in implementing recommendation systems in libraries has also become a hot topic of discussion in the modern library world. To create a recommendation system using machine learning, set data training and testing is needed which can be obtained from implicit or explicit feedback. However, generally what libraries have is implicit feedback in the form of borrowing history data, rather than explicit feedback such as reviews or user surveys. In this study, we propose to use Alternating Least Squares (ALS) in creating recommendation systems in libraries by utilizing borrowing history data. Book loan record was retrieved from University library database for two years period resulting 17,328 book loan transaction record. Filtered to get data of active student who transaction more than 1 book per month, resulting in 1041 book loan record. This data then processed in ALS model using PySpark lib in Python using google collab notebook. ALS, suitable for large-scale collaborative filtering tasks with implicit data. The matrix used is a collaboration of users and items, users here are library users and items are library collection books. The result of the model is a recommendation system which is a prediction of collections that might be borrowed by users based on ALS modeling. The performance evaluation of this system recommendation model will be shown from the results of precision, recall, and F1-score to measure the accuracy of recommendations and Root Mean Square Error (RMSE) and Mean Absolute Error (MAE) to measure the quality of predictions.