Interpretable heart disease risk prediction via FCA-constrained logistic regression.

Area of Science:

• Machine Learning
• Biostatistics
• Cardiology

Background:

• Heart disease risk prediction models are crucial for early intervention.
• Existing models may lack interpretability or clinical coherence.
• Integrating structured data analysis with predictive modeling is an active research area.

Purpose of the Study:

• To develop an interpretable and clinically coherent heart disease risk prediction model.
• To integrate Formal Concept Analysis (FCA) with a novel closure-constrained logistic regression.
• To enforce coefficient coherence within FCA-derived concepts for enhanced model understanding.

Main Methods:

• Utilized the Heart Disease Health Indicators dataset (BRFSS 2015; N≈380,000).
• Discretized predictors into binary attributes and extracted closed itemsets via FCA.
• Applied a closure penalty to logistic regression to minimize within-concept coefficient variance.
• Selected hyperparameters using five-fold cross-validation and evaluated on a held-out test set.

Main Results:

• The FCA-constrained model achieved Accuracy = 0.906, AUC = 0.810, Precision = 0.709, F1 = 0.556, and Brier Score = 0.078.
• Outperformed baseline models (L2-LR, RF, GB) in precision and F1-score.
• Produced well-calibrated probabilities and provided concept-level explanations.

Conclusions:

• Embedding FCA structure into model training yields an interpretable linear model.
• The developed model demonstrates competitive discrimination and improved precision.
• This approach offers clinically coherent explanations for heart disease risk predictions.