Two-stage hybrid feature selection algorithms for diagnosing erythemato-squamous diseases

Area of Science:

• Computational biology
• Machine learning in healthcare
• Biomedical informatics

Background:

• Developing stable and efficient diagnostic models is crucial for accurate disease classification.
• Traditional feature selection methods may not always yield optimal results for complex datasets.
• Hybrid approaches combining filter and wrapper methods offer potential for improved model performance.

Purpose of the Study:

• To propose and evaluate novel two-stage hybrid feature selection algorithms for enhanced diagnostic model construction.
• To introduce a new accuracy measure for guiding feature selection processes.
• To improve classification accuracy in diagnosing erythemato-squamous diseases.

Main Methods:

• Utilized Support Vector Machines (SVM) as the core classification tool.
• Employed Sequential Forward Search (SFS), Sequential Forward Floating Search (SFFS), and Sequential Backward Floating Search (SBFS) as search strategies.
• Incorporated the generalized F-score (GF) for feature importance evaluation and a novel accuracy measure for model assessment.
• Implemented a two-stage process involving 10-fold cross-validation and a second-stage selection on the best-performing fold.

Main Results:

• The proposed two-stage hybrid feature selection algorithms constructed more efficient diagnostic models compared to single-stage methods.
• Experimental results demonstrated superior classification accuracy for the developed models.
• The novel methods outperformed existing algorithms in diagnosing erythemato-squamous diseases.

Conclusions:

• The two-stage hybrid feature selection approach effectively enhances the stability and efficiency of diagnostic models.
• The introduced accuracy measure serves as a reliable criterion for directing feature selection.
• This methodology offers a promising advancement for accurate disease diagnosis, particularly for erythemato-squamous conditions.