Multiple feature selection based on an optimization strategy for causal analysis of health data

Area of Science:

• Health Informatics
• Data Science
• Biostatistics

Background:

• Advancements in information technology and wearable devices generate complex health datasets.
• Causal graphs are crucial for understanding health feature relationships but face computational challenges.
• Feature selection is vital for managing high-dimensional health data and improving analytical efficiency.

Purpose of the Study:

• To present a framework for multiple feature selection using an optimization strategy for causal analysis of health data.
• To address challenges in health analytics, including large feature sets and computational demands.
• To enhance the identification of significant relationships within complex health data.

Main Methods:

• A Weighted Total Score (WTS) index was defined to assess combined feature importance from multiple selection methods.
• A greedy algorithm was integrated to optimize weights for each feature selection method.
• Causal graphs were constructed using selected features, with statistical significance assessed for causal paths.

Main Results:

• The proposed framework reduced the number of features while improving model performance compared to baseline models.
• The statistical significance of feature relationships identified through causal graphs was validated.
• The approach demonstrated effectiveness on both a custom dataset and an open diabetes dataset.

Conclusions:

• The developed framework successfully reduces feature dimensionality in health data analysis.
• It effectively uncovers and validates causal relationships among health features.
• This contributes to improved healthcare and public health strategies through advanced data analysis.