Feature sensitivity criterion-based sampling strategy from the Optimization based on Phylogram Analysis (Fs-OPA) and...

Area of Science:

• Health Informatics
• Data Science
• Machine Learning

Background:

• Healthcare facilities accumulate vast digital patient data over time.
• Analyzing these large datasets is challenging due to a lack of specialized expertise and high costs.
• Existing methods often require domain-specific knowledge, limiting their application.

Purpose of the Study:

• To introduce a novel technique, Feature Sensitivity-Optimization based on Phylogram Analysis (FS-opa), for automatic analysis of large healthcare datasets.
• To enable the identification of principal features for predictive modeling without relying on domain experts.
• To improve the efficiency of healthcare data analysis and model building.

Main Methods:

• FS-opa employs a criterion-based sampling strategy derived from Optimization based on Phylogram Analysis.
• The method processes raw data, mining entire datasets from scratch.
• It integrates with statistical or machine learning methods for predictions and uses Cox's approach for survival analysis.

Main Results:

• FS-opa successfully performed feature sensitivity analysis on raw electronic health record data from mental disorder prehospital services.
• The technique identified principal features without bias from inference models.
• The generated models demonstrated a beneficial trade-off between representativeness and parsimony.

Conclusions:

• FS-opa offers an automated solution for analyzing large, complex healthcare datasets.
• The technique reduces the need for domain expertise in feature selection for predictive modeling.
• FS-opa can enhance expert analysis by highlighting key aspects for improved hospital efficiency and patient treatment quality.