Great strides have been made in machine learning to make models more interpretable and hence more suitable to complex domains such as healthcare [source]. Crucial to the adaptation of these models in industry is the creation of tools that allow any user to ask questions of these models.

Working closely with other researchers, I was given the task of taking raw model output and creating a visualization platform that could answer specific patient level information but also provide high-level cohort level information. At the core of the project was a focus on interoperability between high level insights and low level raw data. With just several interactions a user can pick a cluster of patients with the highest occurrence of a given disease, and from that cluster find the patients with the highest risk score for heart failure. Once a patient is selected their timeline can be zoomed to view events for only several days.

Screen Shot 2016-09-05 at 8.04.43 PM

Tools used: Python (numPy, SKlearn) for data processing and machine learning, JavaScript (D3, jQuery) for data visualization

My Role: I was responsible for data processing, clustering, UI design and creation.

note: unfortunately, because of the sensitivity of patient data, no working demo can be provided. Please contact me if you would like more information on the project.