Iterative regression ensembles for spatio-temporal forecasting
Iterative regression ensembles for spatio-temporal forecasting
Lior Kaspa, Tel-Aviv University Advisor: : Prof. Neta Rabin (TAU)
Abstract:
Space and time are ubiquitous aspects of observations in several domains, including climate science, neuroscience, social sciences, epidemiology, transportation, criminology, and Earth sciences, which are rapidly being transformed by the deluge of data. Spatiotemporal models arise when data are collected across time as well as space and have at least one spatial and one temporal property. Typically, spatio-temporal forecasting is done either by statistical methods, machine or deep learning models. Each have their limitation, some related to the large parameters of the model that need to be determined and others due to their blackbox nature. This work focuses on kernel-based regression techniques, which provide a non-parametric regression model. Kernel-based regression methods can tackle challenging tasks that introduce non-linear or non-stationary data. In previous work, kernel based iterative regressions were extended to handle spatiotemporal data. However, the model required carefully setting the parameters that capture the spatial dependencies. In this work we proposed to build an ensemble-based regression technique, which is also robust and can handle missing values. Experimental results are tested on real world datasets and compared with machine learning models.
Bio:
Lior Kaspa, M.Sc. student at the Department of Industrial Engineering at Tel Aviv University, specializing in Business Analytics under the supervision of Prof. Neta Rabin. Lior also holds a B.sc degree in Industrial Engineering and management sciences from Tel Aviv University. Lior is a team leader in the IDF at Lotem – the IDF digital and data unit, and is responsible for various software development projects.models
Contact:
• E-Mail: liorkaspa@mail.tau.ac.il