Deep Learning Layers to increase training accuracy

Hi everyones.

I want to use DL for modeling a problem with 6 inputs and one output.

I've used the following layers but I cannot increase the model accuracy not only for unseen samples but also for training samples.

I've check different structures and try to generated a complex model as much as possible to get at least good results in training stages

numHiddenNeuron = 100;

layers = [

featureInputLayer(numFeatures,'Normalization','rescale-symmetric')

fullyConnectedLayer(numHiddenNeuron)

reluLayer('Name','relu')

batchNormalizationLayer

fullyConnectedLayer(numOut)

regressionLayer('Name','regression')];

I would be appreciated it if you could help me.

Regards,

Answer 1

[yanqi liu]

yes，sir，may be add some dropoutLayer in net Layers，such as

numHiddenNeuron = 100;
layers = [
            featureInputLayer(numFeatures,'Normalization','rescale-symmetric')
            fullyConnectedLayer(numHiddenNeuron)
            reluLayer('Name','relu')
            batchNormalizationLayer
            dropoutLayer
            fullyConnectedLayer(numOut)
            regressionLayer('Name','regression')];

% or
layers = [ ...
    sequenceInputLayer(numFeatures)
    lstmLayer(100,'OutputMode','sequence')
    dropoutLayer(0.3)
    lstmLayer(50,'OutputMode','sequence')
    dropoutLayer(0.2)
    fullyConnectedLayer(numOut)
    regressionLayer];

if possible，may be upload your data to analysis

[Jahetbe]

Thank you for your response.

I've check the first one and the results didn't change.

For the 2nd one, my data is not time series and I cannot use the "sequenceInputLayer" as well as "lstmLayer".

Dou you have any other recommendation? Infortunately the resutls are vey bad

Regards

[Jahetbe]

I've check it agsin, unfortunately after some iterations (lest than 20), the RMSE does not decrease