icarm

Interpretable Contextual-Accountable and Responsible Machine Learning

A general-purpose framework for Interpretable Contextual-Accountable and Responsible Machine Learning (ICARM) that works with any clean tabular data across any application domain including healthcare, finance, social science, business, and education. Automatically detects whether a prediction task is binary classification, multi-class classification, or regression from the target variable type. Provides a unified entry point icarm_fit() supporting both interpretable learners (Classification and Regression Trees (CART), logistic regression, linear regression, Generalized Additive Models (GAM)) and extended learners (random forest, 'XGBoost', Support Vector Machines (SVM)) with consistent interfaces for global and local model explanation, group-level fairness auditing across protected attributes, probability calibration, threshold analysis, multi-model comparison, reproducible JavaScript Object Notation (JSON) audit trails, and accountability scorecards. The contextual accountability framing emphasises that algorithmic fairness and interpretability requirements depend on the deployment domain and must be evaluated accordingly. Extends the 'civic.icarm' framework (Awe 2025) <https://cran.r-project.org/package=civic.icarm> to general-purpose applications beyond civic and political education.

Versions across snapshots

Dependencies (latest)