/FEHCClassifier

Primary LanguagePython

Forest of Evolutionary Hierarchical Classifiers (FEHC)

  • Code for the manuscript "To Combat Multi-class Imbalanced Problems by Aggregating Evolutionary Hierarchical Classifiers" (In Submission)

  • Required Python 3 packages:

    1. sklearn (https://github.com/scikit-learn/scikit-learn)
    2. imblearn (https://github.com/scikit-learn-contrib/imbalanced-learn)
    3. joblib (for dataset loading, datasets = joblib.load('MCIDatasets.pkl'))

  • FEHC is compatible with most sklearn APIs but is not strictly tested.

  • Import: from FEHCClassifier import FEHCClassifier

  • Train: fit(X, y), with target $y_i \in {0, ..., K - 1}$ as the labels.

  • Predict: predict(X) (hard prediction), predict_proba(X) (probabilistic prediction), or predict(X, n_estimator=1) (using the EHMC instead of ESAE to predict, faster but possibly leading to performance degradation).

  • Non-trivital parameters:

    1. base_estimators: dict, default={'DT': DecisionTreeClassifier()}, candidate classifier set $\mathcal{C}$, should have predict_proba() function"
    2. n_estimators: int, default=30, "the number of EHMCs $k$ in the FEHC"
    3. population: int, default=10, "the population size $\theta_P$ of the MCGA"
    4. iteration: int, default=5, "the number of iteration rounds $\theta_I$ of the MCGA"

  • Related multi-class imbalanced datasets can be found in Releases.