Random forests for functional covariates

Abstract

We propose a form of random forests that is especially suited for functional covariates. The method is based on partitioning the functions' domain in intervals and using the functions' mean values across those intervals as predictors in regression or classification trees. This approach appears to be more intuitive to applied researchers than usual methods for functional data, while also performing very well in terms of prediction accuracy. The intervals are obtained from randomly drawn, exponentially distributed waiting times. We apply our method to data from Raman spectra on boar meat as well as near‐infrared absorption spectra. The predictive performance of the proposed functional random forests is compared with commonly used parametric and nonparametric functional methods and with a nonfunctional random forest using the single measurements of the curve as covariates. Further, we present a functional variable importance measure, yielding information about the relevance of the different parts of the predictor curves. Our variable importance curve is much smoother and hence easier to interpret than the one obtained from nonfunctional random forests.

Modeling of Aboveground Biomass with Landsat 8 OLI and Machine Learning in Temperate Forests

“…The study corroborates the potential of machine-learning methods for AGB prediction and mapping as successfully utilized in recent studies (e.g., [49,51,[77][78][79][80][81][82][83]). In our study, the performance of SVR was better than RF.…”

“…The training of the RF model consisted on optimizing the parameters ntree (number of trees grown) and mtry (number of predictors sampled for spliting at each node, to minimize the influence of a very strong predictor against the other variables) [77]. The parameters mtry and ntree were optimized by evaluating their effect on the mean quadratic error.…”

Mathematical optimization in classification and regression trees

“…We end the section with some considerations on other types of data, apart from continuous and binary, available to construct tree models that show a good trade-off between accuracy and sparsity. When building classification and regression models, there may be characteristics which are recorded as, for example, time-series data ( Barrow and Crone 2016;Saha et al 2020), spatial data (Georganos et al 2019), functional data (Balakrishnan and Madigan 2006;Möller et al 2016;Pospisil and Lee 2019;Rahman et al 2019), text data (Martens and Provost minimize ‖ ‖ s.t. Π k * (x 0 + ) ≥ Π k (x 0 + ) ∀k = 1, … , K ∈ A.…”

A Two-Sample Distribution-Free Test for Functional Data with Application to a Diffusion Tensor Imaging Study of Multiple Sclerosis

“…One example was studied by Möller et al . (), where the interest was to test whether the Raman spectra of boar‐tainted or non boar‐tainted mean samples have the same distribution. Another example is a recent study of activity in cats suffering from degenerative joint disease where the interest is to test whether the minute‐by‐minute activity profiles of the cats which receive treatment vary differently compared with those which receive placebo.…”