An RBF neural network based on improved black widow optimization algorithm for classification and regression...

Area of Science:

• Machine Learning
• Artificial Intelligence
• Computational Science

Background:

• Regression and classification are foundational machine learning tasks.
• Developing robust models with enhanced generalization is crucial for complex problems.

Purpose of the Study:

• To introduce the improved black widow optimization algorithm (IBWO) for radial basis function neural networks (RBFNNs).
• To enhance the generalization capability of RBFNNs using a nonlinear time-varying inertia weight.
• To evaluate the proposed IBWO-RBF model on various classification and regression benchmarks.

Main Methods:

• Development of the IBWO-RBF model integrating an improved black widow optimization algorithm with RBFNNs.
• Implementation of a nonlinear time-varying inertia weight to boost model generalization.
• Application of the IBWO-RBF model to UCI dataset classification, nonlinear function approximation, system identification, and power load prediction.

Main Results:

• The IBWO-RBF model achieved high accuracy and parsimony across diverse classification and regression problems.
• Experimental validation confirmed the model's effectiveness on benchmark datasets and a practical power load prediction task.
• The proposed model outperformed existing methods in terms of both predictive accuracy and model simplicity.

Conclusions:

• The IBWO-RBF model offers a powerful and efficient solution for complex machine learning classification and regression tasks.
• The integration of IBWO with RBFNNs, particularly with the nonlinear time-varying inertia weight, significantly improves generalization.
• The model's success in real-world applications like power load prediction highlights its practical utility and robustness.