Novel Time Series Machine Learning Methodology for High-Dimensional Data

This PhD project focuses on developing novel time series machine learning methodology for high-dimensional data, with emphasis on missing value imputation and accurate forecasting. The research is especially focused on difficult cases involving high-dimensional and discrete-valued data, where existing methods remain limited. The project has two main research areas: Imputation of missing data in high-dimensional time series modelling network relationships between components combining temporal dependency with state-space methods applying dimension reduction through factor models developing self-exciting spatio-temporal models handling both continuous and discrete-valued data Machine learning architectures for robust forecasting developing deep learning models based on temporal convolutional networks and transformers extending forecasting methods to high-dimensional and discrete-valued settings combining factor models with probabilistic and statistical hybrid methods improving uncertainty quantification using Bayesian inference and quantile regression The project will also validate the methods using real-world case studies in: finance healthcare environmental monitoring Expected outputs include journal publications, open-source AI models, and deployment-ready prototypes.

Strong background in mathematics, statistics, econometrics, machine learning, data science, or related field
Interest in time series modelling, forecasting, and missing data methods
Strong quantitative and analytical skills

Preferred:

Interest in deep learning
Interest in probabilistic modelling
Experience with high-dimensional dataset

Contact the supervisors:
Dr Jiazhu Pan
Prof Ke Chen
Apply through the University of Strathclyde PhD application route
Include the project title and supervisor details in the application