Dimensionality Reduction in Complex Medical Data: Improved Self-Adaptive Niche Genetic Algorithm.

Area of Science:

• Biomedical Informatics
• Computational Biology
• Medical Data Analysis

Background:

• High-dimensional clinical datasets pose challenges for classification and model efficiency.
• Conventional Niche Genetic Algorithms (NGA) have limitations due to fixed niche distance parameters.
• Effective dimensionality reduction is crucial for accurate clinical data analysis.

Purpose of the Study:

• To introduce an Improved Niche Genetic Algorithm (INGA) for enhanced dimensionality reduction in clinical datasets.
• To improve population diversity and prevent local optima in genetic algorithms.
• To validate the efficacy of INGA in a sepsis patient stratification model.

Main Methods:

• Developed an Improved Niche Genetic Algorithm (INGA) with a self-adaptive niche-culling operation.
• Applied INGA to reduce feature dimensionality in high-dimensional clinical data.
• Utilized INGA for a sepsis patient stratification model to predict 28-day mortality.

Main Results:

• Reduced feature dimensionality from 77 to 10 using INGA.
• Achieved 92% accuracy in predicting 28-day death in sepsis patients.
• Demonstrated superior performance compared to other methods in sepsis patient stratification.

Conclusions:

• INGA effectively reduces dimensionality in complex clinical datasets.
• The self-adaptive niche-culling in INGA improves model performance and prevents local optima.
• INGA shows significant potential for improving clinical decision-making, particularly in sepsis management.