How to retain a program that has be trained by Radial Basis Function?

Hello every one i made and trained a program using the nntool, i used the Radial Basis Function as a training algorithim. the results were good but i want them with better accuracy, is there any way to change the settings of the fucntion to give better answers, also can i get some details of the trained program such as: no. of iterations, error plot and etc. thank you

Answer 1

[Krishna]

Hi Mohammed,

You can improve the accuracy of your RBF network by adjusting some key settings. One of the most important factors is the spread parameter, which controls the size of the influence of each RBF neuron. A smaller spread makes the neurons sensitive to smaller areas, while a larger spread makes them cover a broader region. Experimenting with different spread values can help improve performance. Additionally, you might want to adjust the learning rate if you're using a gradient descent-based optimization technique. If you feel that the model is not capturing enough complexity, increasing the number of neurons in the hidden layer could offer better results, as it provides more flexibility in approximating the data. Regularization techniques can also help if you're facing overfitting.

To gain insights into your trained network, you can access various details in MATLAB. To see the number of iterations during training, use the 'trainParam' structure from the trained network object. You can also track the error during training using 'trainErrorFcn', and use the 'plotperform(net)' function to visualize the training performance curve. For further training details such as training time and final error, you can access properties like 'trainFcn' and 'trainParam.'

These adjustments and insights will help you refine your model for better performance and provide more visibility into the training process.

Hope this helps.