easeai 0.1.0

Reusable XGBoost + SHAP workflow for tabular regression explainability

EASEai

Explainable AI for epidemiology and public health.

easeai is a lightweight Python package for tabular regression explainability built around a practical research workflow:

• XGBoost hyperparameter tuning
• iterative feature elimination based on XGBoost importance
• cross-validated RMSE and R²
• SHAP feature ranking
• partial dependence plots
• row-level dominant driver extraction

It was extracted from a county-level environmental health workflow, but the package is intentionally general so others can use it on any tabular regression dataset.

Why use EASEai?

Many research notebooks mix together preprocessing, model tuning, feature selection, evaluation, and explainability in one place. easeai turns that into a reusable package for researchers who want:

• a quick XGBoost + SHAP baseline
• interpretable feature ranking
• reproducible artifact export
• a cleaner starting point for GitHub or publication-oriented workflows

Installation

pip install -e .

Or after publication:

pip install easeai

Minimal example

import pandas as pd
import easeai as ea


df = pd.read_csv("Alzheimer_merged1.csv", encoding="ISO-8859-1")

workflow = ea.TabularXAIRegressor(
    target="AD_PREV_MEAN",
    drop_columns=["Counties", "FIPS"],
    target_n_features=15,
)

workflow.fit(df)
results = workflow.summarize(id_column="FIPS", name_column="Counties")

print(results.selected_features)
print(results.metrics)
print(results.shap_importance.head())

results.top_drivers[["id", "name", "top_driver"]].to_csv("county_top_drivers.csv", index=False)
workflow.export_artifacts("artifacts")

Package structure

easeai/
  data.py        # preprocessing helpers
  model.py       # tuning, CV, recursive elimination
  explain.py     # SHAP summaries and top-driver extraction
  plotting.py    # SHAP and PDP export helpers
  workflow.py    # end-to-end workflow class

Suggested GitHub topics

xgboost, shap, explainable-ai, tabular-data, epidemiology, public-health, machine-learning, python

Roadmap

• add classification support
• add permutation importance
• add bootstrap confidence intervals
• add optional map-ready exports
• publish on PyPI

Development

pytest

License

MIT

Author

Ria A. Martins