Spatial fisheries management needs accurate spatial data on fish distribution and fishing impacts. In the NW Mediterranean, small pelagic fish stocks are monitored with acoustic surveys, but these lack high temporal resolution. This study used machine learning (random forest) to classify purse seiner vessel activities from VMS data into fishing, tracking, and cruising. The model improved prediction accuracy and provided more reliable fishing effort estimates than traditional speed-filter methods, helping improve spatial CPUE assessments and fisheries management.