Data calibration in ANN

Which are the different calibration techniques in ANN that is used in basic ann forecasting model?

Any example files is there to refer?

[NN]

How can i calibrate the data in nntraintool?

Below is the code that i am using.

% Setup Division of Data for Training, Validation, Testing

net.divideParam.trainRatio = 65/100;

net.divideParam.valRatio = 5/100;

net.divideParam.testRatio = 30/100;

Answer 1

[Neha]

Hi,

I understand that you want to know about the data calibration techniques used in ANN. You can refer to the following common calibration (or data preprocessing) techniques used in ANN:

Normalization: This technique scales the data to fit within a certain range, usually 0 to 1 or -1 to 1. You can use "mapminmax" function to normalize your data. You can refer to the following documentation link for more information on "mapminmax":

https://www.mathworks.com/help/deeplearning/ref/mapminmax.html

However if you are using deep learning workflows, you can normalize the data in the input layer itself (Eg: "sequenceInputLayer", "imageInputLayer", "featureInputLayer", etc) using normalization name-value pair instead of using "mapminmax".

Standardization: This technique transforms the data to have zero mean and unit variance. This is useful when the data follows a Gaussian distribution. You can use "mapstd" function to standardize the data. You can refer to the following documentation link for more information on "mapstd":

https://www.mathworks.com/help/deeplearning/ref/mapstd.html

Apart from the above two techniques, if your data is categorical, you can use the "onehotencode" function to create a binary column for each category and mark it with a 1 for the corresponding category. You can refer to the following documentation link for more information on "onehotencode":

https://www.mathworks.com/help/deeplearning/ref/onehotencode.html

You have also mentioned that you have used "nntraintool" for neural network training, since this function is deprecated, you can use the "train" function instead to train shallow neural networks:

https://www.mathworks.com/help/deeplearning/ref/network.train.html

Hope this helps!