Regularised Estimation and Machine Learning in Spatiotemporal Statistics

 

Researchers in the school investigate how regularisation and machine learning can improve inference and prediction in spatiotemporal models. Projects span adaptive LASSO, cross-validation methods for dependent data, and model-based neural networks embedded in distributional regression frameworks. A key theme is balancing flexibility and interpretability when modelling large-scale environmental or financial systems. Recent work includes benchmarking computational tools for regularised spatiotemporal models and developing hybrid frameworks that combine deep learning architectures with structured statistical predictors. These methods are applied to domains such as air pollution forecasting, urban mobility, and additive manufacturing.

Researchers

Publications

  • Otto, P., Fassò, A., Maranzano, P. (2024): A review of regularised estimation methods and cross-validation in spatiotemporal statistics. Statistics Surveys. → Foundational synthesis of regularisation methods in spatial contexts.
  • Otto, P., Schmid, W. (2022): A general framework for spatial GARCH models. Statistical Papers. → Includes regularisation structure and flexible estimation design.
  • Malinovskaya, A., Mozharovskyi, P., Otto, P. (2024): Statistical process monitoring of artificial neural networks. Technometrics. → Connects regularisation, interpretability, and deep learning pipelines.
  • Merk, M., Otto, P. (2022): Estimation of the spatial weighting matrix for regular lattice data – An adaptive lasso approach. Environmetrics. → Targeted contribution on penalised estimation of spatial structure.
  • Otto, P., Steinert, R. (2023): Estimation of the Spatial Weighting Matrix under Structural Breaks. Journal of Computational and Graphical Statistics. → Addresses regularised estimation with change-point detection.