How to remove outlier from training data before training artificial neural networks ?

If there is any method to do this in Matlab kindly share ?

Answer 1

[Greg Heath]

Use zscore or mapstd to transform to zero-mean/unit-variance.

Plot

Decide how many standard deviations from the mean to set the outlier threshold.

Hope this helps.

Thank you for formally accepting my answer

Greg

[Staffan]

Hello Greg,

Applying a threshold “zscored values” would only remove outliers around peaks and troughs, could a filtering technique or putting a threshold on the absolute value of the derivative of the signal be applicable when removing outliers?

Regards

Staffan

[Greg Heath]

Those are not ouliers.

If you have a mathematical or physical reason for removing derivative outliers, compare with a model without this adjustment.

Then decide.

Hope this helps.

Greg

Answer 2

[Image Analyst]

See this from a Mathworker, Brett Shoelson: http://www.mathworks.com/matlabcentral/fileexchange/3961

Also, MAD can be used: https://en.wikipedia.org/wiki/Median_absolute_deviation, attached.