Table 13 The adopted approaches for attaining ethical aspects of AI in this research problem.
From: Stress management with HRV following AI, semantic ontology, genetic algorithm and tree explainer
Ethical AI considerations | Adopted approaches |
|---|---|
Collection of ethical data | We have used a publicly available anonymous HRV dataset for the experiment. |
Test for bias | The use of calibrated classification, stratification, grid search, and metrics, such as MCC, to test for bias in our models. |
Test for fairness | The use of SMOTE and ADASYN algorithms for data balancing and a comparative analysis between balanced and imbalanced datasets. |
Model Explainability | The use of ML-based local model Explanation with Tree-based approach (e.g., SHAP). |
Ethical principles | The testing for bias, fairness, model scalability, and explainability |
Privacy consideration | Data balancing with sampling algorithms, such as SMOTE and ADASYN preserve data privacy. |
Iterate and improve | Incremental modeling for handling growing HRV data, continuous training-validation-testing, and automatic stress detection. |
