icarm

Interpretable, Contextual-Accountable and Responsible Machine Learning for Python

Python port of the icarm R package. Provides a unified pipeline for task-agnostic modelling, explanation, fairness auditing, data drift detection, cross-validation, and accountability reporting.

Installation

pip install icarm

# With extended model support (XGBoost, LightGBM, GAM, SHAP)
pip install "icarm[extended]"

Quick start

from icarm import IcarmModel, icarm_split, load_medical

# Load built-in dataset
df = load_medical()
X  = df.drop(columns=["readmitted"])
y  = df["readmitted"]

# Split
(X_tr, y_tr), (X_te, y_te) = icarm_split(X, y, prop=0.75)

# Fit (task auto-detected: binary)
model = IcarmModel(model="logistic", positive="Yes")
model.fit(X_tr, y_tr)
print(model)

# Performance
model.metrics(X_te, y_te)

# Global explanation
ex = model.explain()
ex.plot()

# SHAP values
shap = model.shap(X_te, n_samples=50)
shap.plot()

# ALE profile
ale = model.ale(X_te, feature="num_prior_visits")
ale.plot()

# Fairness audit
f = model.fairness(X_te, y_te, protected="gender")
f.plot(metric="dp_ratio", ref_line=0.8)
f.summary()

# Accountability
model.scorecard(analyst="O. O. Awe",
                project="Hospital Readmission Study")
trail = model.audit(analyst="O. O. Awe")

Functional API (mirrors R)

from icarm import (
    icarm_fit, icarm_split, icarm_metrics,
    icarm_explain, icarm_shap, icarm_ale,
    icarm_fairness, icarm_equity_summary,
    icarm_calibrate, icarm_thresholds,
    icarm_drift, icarm_cv, icarm_compare,
    icarm_audit, icarm_scorecard,
)

m = icarm_fit(X_tr, y_tr, model="logistic", positive="Yes")
icarm_metrics(y_te, m.predict(X_te),
              y_prob=m.predict(X_te, type="prob")["Yes"].values,
              positive="Yes")

Supported models (18 total)

R package

The companion R package is available on CRAN:

install.packages("icarm")

Author

Olushina Olawale Awe Alexander von Humboldt Visiting Professor, PH Ludwigsburg, Germany ORCID: 0000-0002-0442-4519

Citation

@software{Awe_icarm_python_2026,
  author  = {Awe, Olushina Olawale},
  title   = {{icarm}: Interpretable, Contextual-Accountable and
             Responsible Machine Learning for Python},
  year    = {2026},
  url     = {https://github.com/Olawaleawe/icarm-python},
  version = {0.1.0}
}

License

MIT