Functional Linear Regression Analysis

Functional linear regression analysis extends classical regression to contexts where predictors, responses or both are functions over continuous domains. In scalar-on-function models, a scalar outcome is linked to one or more functional covariates via an integral operator involving an unknown coefficient function. In function-on-function models, the response itself is functional, yielding a surface or bivariate coefficient mapping input curves to output curves. Estimation typically employs basis expansions—such as spline or wavelet bases—combined with smoothing penalties that control curvature and prevent overfitting. Alternatives draw on reproducing kernel Hilbert spaces to manage complexity in high or growing dimensional settings. Key challenges include ensuring identifiability of the coefficient function, selecting tuning parameters, and efficiently computing estimates in the presence of noise, heteroscedasticity or outliers. Applications span biomedicine, environmental monitoring, finance and engineering, where densely observed time-series or spatial profiles naturally invite functional treatment. Recent advances address scalability in partially functional models, robust estimation in the presence of anomalies, and diagnostic tools for assessing model fit and inference reliability.

Research from Nature Portfolio

No recent Nature Portfolio content available.

Research from all publishers

Recent work on high-dimensional partially functional linear models characterises non-asymptotic bounds for prediction risk when the number of scalar covariates grows with the sample size. By leveraging kernel principal components and sandwich operator spectral properties, the approach delineates a trade-off between effective functional dimension and multivariate predictor load, ensuring minimax optimality and consistency under sub-Gaussian conditions. Another line of research proposes a test for conditional independence in Hilbert spaces, employing residual inner products derived from ridge-based functional linear regressions. This method controls type I error without restrictive eigen-spacing assumptions and facilitates inference on truncation points in truncated functional linear models as well as edge detection in functional graphical models. In applied settings, multiple function-on-function principal component regression has been adapted for COVID-19 data imputation, using principal component scores of fully observed curves to estimate missing hospitalisation and intensive care trajectories. Canonical correlation analysis of resulting components further interprets the interplay between health outcomes and covariate curves, demonstrating practical impact in epidemiological studies.

Functional Linear Regression Analysis publication trend

The graph below shows the total number of articles in functional linear regression analysis across all publications each year (not limited to Nature Index journals).

Technical terms

Functional data: Observations represented as curves or functions over a continuous domain, such as time or space.

Scalar-on-function regression: A model linking a scalar response to functional covariates via an integral of the product of covariate and coefficient functions.

Function-on-function regression: A model mapping input functions to output functions through a bivariate coefficient surface.

Basis expansion: Representation of functions using a finite set of known basis functions (e.g. splines, wavelets) to reduce infinite-dimensional problems to finite-dimensional estimation.

Smoothing penalty: An additional term in the estimation objective that penalises roughness of the coefficient function, promoting smooth solutions.

Reproducing kernel Hilbert space (RKHS): A function space defined by a kernel function, enabling regularised estimation and theoretical analysis of functional models.

Functional principal component analysis (FPCA): A dimension-reduction technique decomposing functional data into orthogonal modes of variation, used for representation and estimation.

References

1. Conditional Independence Testing in Hilbert Spaces with Applications to Functional Data Analysis. Journal of the Royal Statistical Society Series B Statistical Methodology (2022).
2. Growing-dimensional partially functional linear models: non-asymptotic optimal prediction error. Physica Scripta (2023).
3. COVID-19 Data Imputation by Multiple Function-on-Function Principal Component Regression. Mathematics (2021).
4. Robust Function-on-Function Regression. Technometrics (2020).