Strategies for penalised least squares estimators of functional regression models

Strategies for penalised least squares estimators of functional regression models

In functional data analysis, the discrete observed data are converted to smooth functions and so they become infinite dimensional data objects.

The analysis involves representing the functional data using a basis expansion and then truncating the basis in term of a finite number of basis elements. Choosing the number of basis elements is part of the data analysis.

Therefore, the dimension of the basis expansion is an unknown parameter and investigation is required to determine its value. A recursive numerical method is examined for choosing the number of basis elements within the context of model selection. Penalised least squares and cross validation procedures are used in order to choose the number of basis elements that optimise the estimation of the functional regression model. The proposed numerical method is based on orthogonal and hyperbolic transformations.

Speaker

Dr Stella Hadjiantoni is a Lecturer in Data Science and Statistics in the Department of Mathematical Sciences.