Sutarman, Adli Abdillah Nababan, Pandi Barita Nauli Simangunsong
This study proposes a hybrid framework for high-dimensional and multimodal data analysis, combining Kernel Principal Component Analysis (KPCA), Kernel Ridge Regression (KRR), and cross-validation with automated hyperparameter tuning via Optuna. The framework is designed to address key challenges including nonlinearity, overfitting, and poor generalization in complex datasets. KPCA enables efficient nonlinear dimensionality reduction, while KRR offers stable and accurate modeling, particularly in multimodal settings. Evaluations across eight empirical and synthetic datasets show that the framework consistently outperforms Support Vector Regression (SVR), achieving low mean squared error (MSE) and mean absolute error (MAE)— with peak performance of MSE = 0.0112 and MAE = 0.0602. The method demonstrates computational efficiency with a training time of less than 80 seconds and memory usage of less than 6 MB. These results underscore the practical value of this framework as a robust and interpretable solution for predictive modeling in high-dimensional data environments. © This article is licensed under a Creative Commons Attribution-ShareAlike 4.0 International License. License details: https://creativecommons.org/licenses/by-sa/4.0/
Department of Mathematics, Faculty of Mathematics and Natural Sciences, Universitas Sumatera Utara, Medan, Indonesia; Information Systems Department, School of Information Systems, Bina Nusantara University, Medan, Indonesia; Department of Data Sciences, Faculty of Computer Sciences, Universitas Katolik Santo Thomas, Medan, Indonesia