A Stability-Aware Hybrid Feature Selection Algorithm Based on Filtered Sparsity and Wrapper Refinement

Key Points

• The aim is to develop a feature selection algorithm that enhances stability and performance in supervised machine learning tasks.
• Developed a hybrid feature selection algorithm combining filtered sparsity with wrapper refinement.
• Applied the algorithm to various datasets of medium to high dimensionality.
• Analyzed the impact of irrelevant and interrelated predictors on model performance.
• The hybrid algorithm showed improved stability in feature selection compared to traditional methods.
• Significant reduction in uninformative features was achieved, enhancing model accuracy.