Pneumonia is a dangerous, often fatal secondary disease acquired by patients during their stay at Intensive Care Units. ICU patients have scores of data collected on a real time basis. Based on two years of data for a large ICU, we develop an early warning system for the onset of pneumonia that is based on Alternating Decision Trees for supervised learning, Sequential Pattern Mining, and the stacking paradigm to combine the two. Mainly due to decreased stay, the system will save € 180000 in this hospital alone while at the same time increasing the quality and consistent standard of health care. The ultimate system relies on a rather small numeric data base alone and is thus amenable to integration in a treatment protocol and a newly conceived ICU workflow system.