Correlation-Based Ensemble Feature Selection Using Bioinspired Algorithms and Classification Using Backpropagation...

Area of Science:

• Computational intelligence
• Medical informatics
• Machine learning

Background:

• Clinical diagnosis relies heavily on accurate data analysis.
• Feature selection is crucial for improving classification model performance.
• Existing methods may not fully leverage bioinspired approaches for complex datasets.

Purpose of the Study:

• To design and implement a novel framework for clinical diagnosis.
• To utilize bioinspired algorithms for optimal feature selection.
• To enhance classification accuracy using a gradient descent backpropagation neural network.

Main Methods:

• Data preprocessing included hot deck imputation and min-max normalization.
• Bioinspired algorithms (Differential Evolution, Lion Optimization, Glowworm Swarm Optimization) were used in a wrapper approach for feature selection.
• Correlation-based ensemble feature selection identified optimal features for training a gradient descent backpropagation neural network.

Main Results:

• The framework achieved 98.47% accuracy on the Wisconsin Diagnostic Breast Cancer dataset.
• An accuracy of 95.51% was obtained for the Hepatitis dataset.
• Ten-fold cross-validation demonstrated robust performance.

Conclusions:

• The proposed framework effectively integrates bioinspired feature selection with neural network classification.
• This approach shows significant potential for developing accurate clinical decision-making systems.
• The framework is adaptable for diagnosing various health disorders.