A Two-Stage Framework for Stable Feature Selection in High-Dimensional Data Using Elastic-Net and Support Vector Machine

High-dimensional data presents a major challenge for classification due to feature redundancy and model instability. To address this, we introduce a novel two-stage framework for stable feature selection. In the first stage, we repeatedly apply logistic regression along with Elastic-Net penalty to rank features based on their selection frequency. This provides a stable and reliable feature ranking, mitigating the inherent instability of single-run methods. The second stage employs an iterative forward-selection approach. We progressively add features to Support Vector Machine (SVM) model, evaluating performance with metrics like AUC, Accuracy, Sensitivity, and Specificity. The process stops when the mean AUC shows no significant improvement. We validate our framework on five high-dimensional gene expression datasets. Our results demonstrate that this method offers superior stability and predictive power compared to traditional feature selection approaches.

کلیدواژه ها:

Feature Ranking ، Elastic-Net ، High-Dimensional Data ، SVM ، Stability Selection