Evolving Hybrid Partial Genetic Algorithm Classification Model for Cost-effective Frailty Screening: Investigative...

Area of Science:

• Gerontology and Artificial Intelligence
• Health Informatics
• Computational Medicine

Background:

• Frailty is commonly measured using a Frailty Index (FI), adaptable to various databases.
• Database structures often hinder direct FI feature extraction, increasing costs.
• Optimizing feature selection is crucial for cost-effective frailty assessment.

Purpose of the Study:

• To apply artificial intelligence (AI) optimization, specifically partial genetic algorithms, to refine feature subsets for FI calculation.
• To prioritize features with lower acquisition costs in the FI calculation process.

Main Methods:

• Secondary analysis of a residential care database (592 residents, aged 75+).
• Utilized a modified genetic algorithm to optimize feature selection for predicting an electronic Frailty Index (eFI).
• Evaluated four classification models (logistic regression, decision trees, random forest, support vector machines) using partial genetic algorithms.

Main Results:

• Logistic regression models performed best across various scenarios and feature set sizes.
• Optimal models combined low-cost features with a minimal number of high-cost features (around 10).
• Achieved high performance metrics (sensitivity 89%, specificity 87%), indicating suitability for low-cost screening.

Conclusions:

• Demonstrated a systematic method for selecting cost-effective features for frailty detection using an aged care database.
• Partial genetic algorithms effectively balance cost and accuracy for systematic frailty identification.
• The optimized approach shows promise for developing low-cost frailty screening tools.