Assessing rbf networks using DELVE

Area of Science:

• Machine Learning
• Artificial Intelligence
• Computational Statistics

Background:

• Radial basis function (RBF) networks are a class of neural networks.
• RBF networks are effective for various regression and classification tasks.
• Optimizing RBF network training is crucial for performance.

Purpose of the Study:

• To describe and illustrate different training methods for RBF networks in regression.
• To empirically compare various RBF training methods against each other.
• To benchmark RBF methods against established non-RBF machine learning techniques using DELVE data.

Main Methods:

• Description and illustration of multiple RBF network training algorithms.
• Empirical evaluation of RBF methods on regression problems.
• Comparative analysis using datasets from the DELVE archive.

Main Results:

• Each RBF training method demonstrated proficiency on specific DELVE tasks.
• No single RBF method consistently outperformed others across all tasks.
• The best-performing non-RBF methods exhibited greater consistency than RBF approaches.

Conclusions:

• RBF networks offer viable solutions for specific regression challenges.
• The consistency of RBF methods may be a limitation compared to other machine learning algorithms.
• Further research into RBF training optimization is warranted for improved generalization.