High dimensional spectral datasets, such as near infrared (NIR) spectra, often suffer from multicollinearity and lack of interpretability when analyzed using conventional regression methods that treat all predictors as a single undifferentiated block. However, few studies have systematically explored how structured, blockwise feature relationships within spectral data can be exploited to enhance prediction accuracy and interpretability. Addressing this research gap, the present study introduces a multiblock feature selection framework based on Elastic Net regression for analyzing high dimensional spectral data from the corn.mat dataset. The dataset comprises NIR spectra divided into five logical wavelength blocks, each independently modeled to predict moisture content. The Elastic Net regularization balances variable selection and multicollinearity handling, while the multiblock structure enables block level comparison and interpretability. Performance metrics, including the coefficient of determination (R 2 ) and Mean Squared Error (MSE), are computed per block to assess predictive relevance. A suite of visualization tools bar charts, dot plots, line graphs, radar plots, and bubble plots illustrates the comparative contribution of each block to overall model performance. Results indicate that certain spectral regions, particularly specific wavelength blocks, contribute disproportionately to prediction accuracy, providing valuable insights for dimensionality reduction and model transparency. The proposed blockwise Elastic Net framework effectively identifies the most informative spectral regions, improving both predictive performance and interpretability. This methodology not only facilitates robust modeling in chemometric and agricultural applications but also offers a scalable analytical tool for complex, multivariate datasets. Beyond this case study, it can be applied to domains such as remote sensing, biomedical signal processing, and multi-omics analysis, where structured, high-dimensional data are prevalent. Ultimately, the study demonstrates how mathematical modeling and multiblock regularization can guide sustainable agricultural regulation and optimize crop quality assessment.
Ashraf et al. (Wed,) studied this question.