A Tandem Evolutionary Algorithm for Identifying Causal Rules from Complex Data.

Area of Science:

• Computational Biology
• Machine Learning
• Bioinformatics

Background:

• Discovering causal rules in complex classification problems is challenging.
• Existing methods may struggle with data heterogeneity, missing values, and imbalanced classes.

Purpose of the Study:

• To develop a novel evolutionary approach for discovering causal rules in complex classification problems using batch data.
• To enhance the accuracy and robustness of causal rule discovery in the presence of various data complexities.

Main Methods:

• Utilizing a hypergeometric probability mass function for principled fitness assessment.
• Employing tandem age-layered evolutionary algorithms to evolve parsimonious archives of conjunctive clauses.
• Implementing separate archive bins for clauses of different orders with dynamic threshold adjustments.

Main Results:

• The proposed method successfully recovered true causal rule sets in synthetic epistatic benchmarks.
• Validation on benchmark problems demonstrated robustness against heterogeneity, epistasis, noise, missing data, and imbalanced classes.
• Successful application to a real-world survey dataset for ecohealth interventions.

Conclusions:

• The evolutionary approach offers a powerful and accurate method for causal rule discovery.
• The technique shows promise for analyzing complex biological and health-related datasets.
• This method can inform interventions for diseases like Chagas disease.