• feature engineering
• healthcare
• intensive care unit
• machine learning
• time series
• waveform data

Machine learning is steadily changing healthcare around the world, from smartwatches monitoring health behavior to NLP applications for automated diagnosis. Within this wide field of research, one promising subfield is represented by bedside applications in hospital’s intensive care units (ICUs). ICUs monitor closely the physiological state of a critical patient and some hospitals have started collecting data and researching for a way to assist the complex task of caring for a critical patient. One of these hospitals is the Sick Children Hospital of Toronto, which is collaborating to this thesis project. 
Together with Sick Children’s doctors we selected a cohort of ICU patients derived from MIT’s MIMIC-III database, singled out the PPG (photoplethysmogram) waveform signal and performed feature engineering in order to apply prediction models for critical events, such as circulatory failure. Other than trying to predict critical events the thesis aims to understand whether feature engineering in the medical context is still useful, given the automated feature extraction capabilities of modern ML techniques.