An Integrated Penalized Regression and Dimensionality Reduction Framework for Robust RNA-Seq Cancer Classification
Qubahan Journal of Social Science · 2026
DOI: 10.48161/qjss.v3n3a17 ↗Kurz gefasst
Fachartikel in der Zeitschrift Qubahan Journal of Social Science, 2026. Verfasst von Fatima Zahra Fakir. Die Arbeit nennt 28 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY-NC-ND 4.0).
Abstract (Originalsprache)
High-dimensional data create major challenges for risk-pattern prediction because predictor counts often exceed sample counts, while noise, multicollinearity, redundant variables, and irrelevant signals can reduce model stability. This study develops and evaluates an advanced statistical modeling framework for classifying cancer-related risk patterns from high-dimensional gene-expression profiles. The framework integrates data auditing, variance-based filtering, supervised feature screening, dimensionality reduction, penalized regression, probability-based classification, and validation. We analyzed a public ribonucleic acid sequencing (RNA-Seq) gene-expression dataset containing 801 samples, 20,531 gene-expression predictors, and five cancer classes. After removing zero-variance predictors and applying structured feature screening, we compared four statistical models: L2-regularized multinomial logistic regression, principal component analysis (PCA)-based logistic regression, least absolute shrinkage and selection operator (LASSO) logistic regression, and elastic-net logistic regression. We assessed model performance using a stratified holdout design and a three-fold validation sensitivity check. The PCA-based logistic regression and L2-regularized multinomial logistic regression models achieved the strongest holdout performance, with 99.38% accuracy, a 99.47% macro F1-score, and a perfect macro area under the receiver operating characteristic curve (ROC-AUC) on the test partition. The PCA-based logistic regression model also produced the lowest log-loss and a low macro Brier score, indicating strong probability separation. LASSO retained a sparse feature subset and showed acceptable stability across folds, supporting interpretability. The findings show that a conservative statistical pipeline can convert high-dimensional molecular data into accurate, interpretable risk-pattern classifications. The study contributes a reproducible framework for balancing predictive accuracy, dimensionality reduction, and model transparency in high-dimensional decision-support contexts.
Zitieren
APA
Fakir, F. Z. (2026). An Integrated Penalized Regression and Dimensionality Reduction Framework for Robust RNA-Seq Cancer Classification. Qubahan Journal of Social Science. https://doi.org/10.48161/qjss.v3n3a17
Vancouver
Fakir FZ. An Integrated Penalized Regression and Dimensionality Reduction Framework for Robust RNA-Seq Cancer Classification. Qubahan Journal of Social Science. 2026. doi:10.48161/qjss.v3n3a17
BibTeX
@article{Fakir2026,
author = {Fakir, Fatima Zahra},
title = {An Integrated Penalized Regression and Dimensionality Reduction Framework for Robust RNA-Seq Cancer Classification},
journal = {Qubahan Journal of Social Science},
year = {2026},
doi = {10.48161/qjss.v3n3a17}
}
EUVIMED