Please use this identifier to cite or link to this item:
http://dx.doi.org/10.25673/123520| Title: | Random Forest-Based Estimation of Partial Linear Single-Index Model in Longitudinal Data |
| Author(s): | Bayyoodh, Hussein Jabbar Aldouri, Mohammed Sadeq |
| Granting Institution: | Hochschule Anhalt |
| Issue Date: | 2026-03 |
| Extent: | 1 Online-Ressource (7 Seiten) |
| Language: | English |
| Abstract: | This paper discusses the estimation of a partial linear single-index model (PLSIM) for longitudinal data. It proposes a hybrid Random Forest-based estimator complemented by statistical regularization to ensure parameter stability and accurate retrieval of the nonlinear component. The proposed approach is based on estimating 𝑔(⋅) via Random Forests with a conservative selection of the number of trees with a 1-SE rule and subject-wise K-fold cross-validation, estimating the trend index 𝛽, and estimating the coefficients of the linear component 𝜃. On balanced simulation data of sizes 𝑁 = [50,100,15] and a real function 𝑔(𝑢) = sin (𝑢), the hybrid estimator showed high accuracy in retrieving 𝑔(⋅) across the sample-supported domain, with a coefficient of determination of 𝑅9 2 ≈ 0.96 − 0.98 and a decreasing mean square error with increasing size, while the overall model performance stabilized at 𝑅2 ≈ 0.90 − 0.92 and 𝑀𝑆𝐸 ≈ 0.10 − 0.13. 𝜃 biases appeared small across all scenarios, while the 𝛽 estimate maintained functional stability reflected in a strong visual match between the real and estimated, with systematically reduced marginal deviations compared to the conventional two-stage estimator, which showed greater sensitivity to the bootstrap parameter and index error and higher overall MSE at the larger sample. The results demonstrate that combining RF with statistical methods provides a practical and accurate path for estimating longitudinal PLSIM models, with straightforward applicability and limited parameter tuning. The study suggests potential future improvements in performance by expanding the framework to account for heteroscedasticity and random effects. |
| URI: | https://opendata.uni-halle.de//handle/1981185920/125454 http://dx.doi.org/10.25673/123520 |
| Open Access: | Open access publication |
| License: | (CC BY-SA 4.0) Creative Commons Attribution ShareAlike 4.0 |
| Appears in Collections: | International Conference on Applied Innovations in IT (ICAIIT) |
Files in This Item:
| File | Description | Size | Format | |
|---|---|---|---|---|
| 1-9-ICAIIT_2026_14(1).pdf | 1.37 MB | Adobe PDF | ![]() View/Open |
Open access publication
