Apache Mahout is a scalable machine learning library built on Hadoop. It provides algorithms for recommendation engines, clustering, classification and other machine learning tasks. Some key algorithms include user-based and item-based collaborative filtering for recommendations, k-means and fuzzy k-means clustering, logistic regression for classification. Mahout is well suited for large datasets and allows machine learning tasks to be easily parallelized across a Hadoop cluster. It has advantages of being open source, scalable, and built on production-quality libraries.