Cluster Boosted Artificial Neural Network (CBANN)

CBANNs enhance ANNs by using additional input variables coming from clustering the training data

Cluster-Boosted Artificial Neural Networks (CBANNs) introduce an innovative approach to improving ANN performance by incorporating cluster identifiers as additional input features. Traditional ANNs often struggle with complex data landscapes, local minima, and nonlinear relationships. CBANNs address these challenges by leveraging unsupervised clustering (e.g., k-medoids) to structure the input space before training.

Key Features of CBANNs:

-Enhanced Pattern Recognition: By adding cluster IDs, CBANNs help the ANN distinguish between different data regions more effectively.

-Improved Accuracy: Compared to conventional ANNs, CBANNs demonstrate significant error reduction on complex benchmark functions.

-Faster Convergence: CBANNs require fewer training epochs to achieve high accuracy, reducing computational cost.

-Broad Applicability: The method has been successfully tested on various benchmark functions (e.g., De Jong’s 5th, Rastrigin, and Griewank) and terrain modeling, showing 95% error reduction in real-world applications.

-Versatile Implementation: CBANNs have been implemented in MATLAB, Python, and Java, with code freely available on GitHub.

CBANNs provide a simple yet powerful enhancement to conventional ANNs, making them particularly useful for applications where high precision and efficient learning are critical.

See the journal paper at Project Website for more details.