Accuracy of results did not increase with training, WHY?

I prepared a fitnet programme. It's working well. But, performance keep on fluctuating always. As my understanding, it should reduce with continuous training. Can you give any suggestions to increase performance?

Answer 1

[Walter Roberson]

Additional training only helps if the fitting model matches the data.

Suppose for example that your function is random. Train on any subset of the random data and you are not going to improve the accuracy of prediction, and you might make the accuracy worse by "overtraining".

[Sugan]

Thank you for your response initially, I'm using 'divideind' function to apply same data for training every time, control the same initial condition by 'stream' function. And '_trainbr_' function used for train the network. But, still accuracy of results are fluctuating. How can i identify it's over trained or not?

[Greg Heath]

Show your code