Multivariate Time Series Analysis with Irregularly Sampled Data

Title Multivariate Time Series Analysis with Irregularly Sampled Data
Summary The student will devise methods for handling irregularly sampled multivariate time series data, addressing missing data and modeling temporal relationships for applications in healthcare
Keywords Machine Learning, Multivariate Time Series, Explainable AI
TimeFrame 2023-2024
Prerequisites Statistics, Neural Networks, Programming (Python or Matlab)
Supervisor Awais Ashfaq, Prayag Tiwari
Level Master
Status Open

The research will commence with a comprehensive review of existing methods, encompassing interpolation techniques and traditional time series models. The student will then explore advanced statistical and machine learning approaches, including dynamic Bayesian networks and deep learning architectures, tailored to irregularly sampled multivariate time series.

The methodologies will be rigorously evaluated using synthetic and real-world datasets, emphasizing predictive accuracy and insights extraction. Potential research directions include:

1. Developing a dedicated imputation method for irregularly sampled multivariate time series. 2. Investigating the application of attention networks, or others for capturing temporal dependencies. 3. Applying the proposed methodologies to real-life healthcare challenges like early detection of diseases. 4. Investigating model interpretability for insights into underlying processes.

Additionally, the student is encouraged to propose and explore their own research questions and directions.

Contact: Awais Ashfaq (