Rotation Forest Classifier

RotationForest is an ensemble learning algorithm designed to improve the accuracy and diversity of decision tree-based classifiers. It was introduced as an extension of the popular RandomForest algorithm. The key idea behind RotationForest is to apply Principal Component Analysis (PCA) to rotate the feature space for each tree in the ensemble, creating diverse and accurate base classifiers.

Unlike RandomForest, which selects a random subset of features at each node, RotationForest:

• Divides features into random subsets and applies PCA transformation to each subset.

• Ensures all original features are used for each tree (instead of random feature selection).

• Uses a C4.5 decision tree (this implementation uses the scikit-learn CART).

Rotation Forest is relevant for Time Series Classification (TSC) because it effectively captures complex feature interactions and correlations which are often critical in time series data using PCA-based rotations. It works well with feature extraction methods (e.g., TSFresh) and is used in TSC pipelines like FreshPRINCE and STC, making it robust for both univariate and multivariate time series data.

In this notebook, we will see how to use the RotationForestClassifier algorithm for time series classification.

# Import necessary libraries
import matplotlib.pyplot as plt
from sklearn.metrics import (
    ConfusionMatrixDisplay,
    accuracy_score,
    classification_report,
    confusion_matrix,
)

from aeon.classification.sklearn import RotationForestClassifier
from aeon.datasets import load_italy_power_demand  # univariate dataset

italy, italy_labels = load_italy_power_demand(split="train")
italy_test, italy_test_labels = load_italy_power_demand(split="test")

italy.shape, italy_labels.shape

((67, 1, 24), (67,))

RotationForestClassifier is not a time series classifier. A valid sklearn input such as a 2d numpy array is required.

# Convert 3D array to 2D array
italy = italy.reshape(italy.shape[0], -1)

italy.shape

(67, 24)

rotation = RotationForestClassifier()
rotation.fit(italy, italy_labels)
y_pred = rotation.predict(italy_test)

accuracy = accuracy_score(italy_test_labels, y_pred)
print("Accuracy: ", accuracy, "\n")

report = classification_report(italy_test_labels, y_pred)
print("Classification Report:\n", report)

Accuracy:  0.9708454810495627

Classification Report:
               precision    recall  f1-score   support

           1       0.97      0.97      0.97       513
           2       0.97      0.97      0.97       516

    accuracy                           0.97      1029
   macro avg       0.97      0.97      0.97      1029
weighted avg       0.97      0.97      0.97      1029

# Plot confusion matrix
cm = confusion_matrix(italy_test_labels, y_pred)
disp = ConfusionMatrixDisplay(confusion_matrix=cm, display_labels=rotation.classes_)
disp.plot(cmap="YlOrRd")
plt.title("Confusion Matrix")
plt.show()

References:

[1] J. J. Rodriguez, L. I. Kuncheva and C. J. Alonso, “Rotation Forest: A New Classifier Ensemble Method,” in IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 28, no. 10, pp. 1619-1630, Oct. 2006, doi: 10.1109/TPAMI.2006.211.

[2] Bagnall, A., Flynn, M., Large, J., Line, J., Bostrom, A., & Cawley, G. (2018). Is rotation forest the best classifier for problems with continuous features? ArXiv. https://arxiv.org/abs/1809.06705

---

Generated using nbsphinx. The Jupyter notebook can be found here.