The project leverages the existing MedSat dataset [1], a comprehensive integration of health data with high-resolution satellite imagery covering 33,000 areas in England. In addition to what is described in the NeurIPS dataset publication, we will be able to offer temporal MedSat dataset, i.e., the version including monthly prescription and environmental indicators for 2 to 5 years. The technical innovation of the proposed project lies in the creation of advanced causal modeling techniques specifically designed to extract meaningful insights from this integrated dataset. Unlike conventional approaches, the project aims to pioneer new methodologies that push the boundaries of what is currently achievable in the domain of causal modeling. Concretely, the student would explore and establish baseline analyses with methods such as [2-4] moving on to exploring more novel approaches [5]. Finally, the research involves the creation of bespoke causal models that go beyond the application of existing techniques.
CausalMED: Novel Causal Methods for Responsible AI in Healthcare
[1] Šćepanović, S., Obadic, I., Joglekar, S., Giustarini, L., Nattero, C., Quercia, D., & Zhu, X. X. (2023, November). MedSat: A Public Health Dataset for England Featuring Medical Prescriptions and Satellite Imagery. In Thirty-seventh Conference on Neural Information Processing Systems Datasets and Benchmarks Track.
[2] Abu Bakar, N., & Rosbi, S. (2017). Autoregressive integrated moving average (ARIMA) model for forecasting cryptocurrency exchange rate in high volatility environment: A new insight of bitcoin transaction. International Journal of Advanced Engineering Research and Science, 4(11), 130-137.
[3] Rosenbaum, P. (2017). Observation and experiment: An introduction to causal inference. Harvard University Press.
[4] Entner, D., & Hoyer, P. O. (2010). On causal discovery from time series data using FCI. Probabilistic graphical models, 121-128.
[5] Zečević, M., Dhami, D. S., Veličković, P., & Kersting, K. (2021). Relating graph neural networks to structural causal models. arXiv preprint arXiv:2109.04173.
MedSat dataset can be downloaded via the following link: https://mediatum.ub.tum.de/1714817.
Project ID
Supervisor
Prof Daniele Querciadaniele.quercia@kcl.ac.ukhttp://researchswinger.org