Dear friends,
I am using RBF neural network to predict the future requests to a web server by a real dataset. The training dataset consists of two-week requests to the server and so do test dataset. Although the test and train set have quite similar patterns, I do not know why the predicted values are so smooth and in a limit range.
I would be grateful if you could help me improve the accuracy of my neural network predictions.
Either training/training data is not sufficient or target values used for training has same range
Make sure the training data set has a wider range than the testing data set. ANN is not that great in extrapolation. I would also randomize the training data set before building the model. You also can try other techniques such as Genetic Expression Tree.
Thank you for your help.
The training data set and testing data set are both from 2 weeks access to a web server.
Nevertheless, I randomized the training data set prior to learning, and it had a fairly useful impact on the neural network prediction accuracy, while its prediction is not still as much accurate as I expected.
- Badges
- Science topic