Prediction of Skin Disease with Three Different Feature Selection Techniques Using Stacking Ensemble Method.

Area of Science:

• Dermatology
• Medical Informatics
• Computational Biology

Background:

• Skin diseases are prevalent, exacerbated by environmental factors like UV radiation and pollution.
• Machine learning and deep learning offer promising avenues for accurate disease diagnosis.
• Erythemato-squamous diseases require effective diagnostic tools for timely intervention.

Purpose of the Study:

• To identify optimal feature subsets for erythemato-squamous disease diagnosis.
• To evaluate the performance of various machine learning classifiers.
• To enhance diagnostic prediction using ensemble techniques.

Main Methods:

• Applied univariate feature selection, feature importance, and correlation matrix with heatmap for feature extraction.
• Utilized Gaussian Naïve Bayesian, Decision Tree, Support Vector Machine, and Random Forest classifiers.
• Implemented a stacking ensemble technique to improve predictive performance.

Main Results:

• Correlation and heatmap feature selection techniques yielded optimal data subsets.
• The stacking ensemble approach demonstrated superior performance compared to individual classifiers.
• Key performance metrics including accuracy, RMSE, and AUC confirmed model effectiveness.

Conclusions:

• Feature selection, particularly using correlation and heatmap methods, is crucial for accurate erythemato-squamous disease diagnosis.
• Stacking ensemble techniques significantly boost prediction performance in skin disease classification.
• The proposed model achieves higher accuracy than previously reported methods, offering a valuable tool for clinical application.