I have proposed RBFNN-PSO model for Prediction of Sediment Transport at the Limit of Deposition in a Clean Pipe. Do RBFNN-PSO model works with Non-normalized Data?
Yes, it certainly works. However, most of the traning codes are designed to work with normilized data.
The wieghts initialization of the activation function in general tries to distribuite the samples in the saturated and non-saturated areas.
In case you do not know the implementation it is better to normalize.
How? the activation function of output unit is sigmoid function, the output with between 0 and 1.
For regression problems it is common to use sigmoid in the hidden layers and linear in the output layer. So, you can get any output.
However, it is easiear and adivisible to normalize. It is not an imposition, but it is a good strategy.
Yes it does. You need to scale the data for drawing any meaningful conclusion.
Yes, by converted into double data type to make it compatible with the ANN Toolbox of Matlab
If you want to be able to make a sensible conclusion you MUST normalize.
This topic has been covered and some nice and interesting answers have already been given on researchgate. You will be able to find them easily if you search the questions/answers with the right keywords (ANN, normalization). Sorry I don't have myself the time to actually do that search and point you to the interesting threads.
Nonetheless, HTH.
Cheers,
SCM
Linear normalization mostly support in a homogeneous type of data set to be input to NN. Furthermore, the normalization will make your algorithm faster. Normalized Radial Functions with similar variances (for input variables) may also allow the NN weights to play their rule more firmly.
Definitely ANN works on non-normalized data but the range of variations in data need not be much high. e.g. you can fuzzify 190, 170, 189, 178, 16, 165 heights of students in cm to tall fuzzy set with trapezoidal membership function without doing the normalization on the given data.
The network can be operated without normalizing the data. However, before entering the signal into the mains lead to the input language network. Any signal can be decomposed into components. Thus each component can be represented as a sequence of single pulses with a repetition rate as a function of the amplitude component. As the results of processing can use the sequence of partial results. When predicting it will provide more adequate evaluation.
The input-output data can be actual or normalized. It is obvious that using normalize data lead to better results but it is not a good idea to use trained model for practical. The range of data is not equal in all problems, then i suggest use actual data
- Badges
- Science topic