Standardized methods for the detection and classification of active sonar have yet to be developed, hindering research into ecological questions related to sonar use over large spatio-temporal scales using archival Passive Acoustic Monitoring (PAM) data. This chapter compares two pipelines for classification of military sonar presence in 20-min files, designed to be generalizable across soundscapes and types of sonar. Pipelines included adapting a deep learning network designed to classify delphinid vocalizations and vessels at the 3-second level to a 20-min resolution using a decision tree, and a Gradient Boosted Random Forest (GBRF) using acoustic indices as features. The adapted deep learning and GBRF pipelines achieved F1 scores of 0.57 and 0.74 respectively. The GBRF pipeline was demonstrated to provide usable predictions of sonar presence, reducing a 51-day dataset to a 12-day period with elevated levels of predicted sonar presence, and 54 out of 935 files predicted to contain sonar predictions outside this period which could be subsequently manually verified. This pipeline is a promising approach for identifying active sonar use in large PAM datasets.