In MATLAB GUI Data set used for making network is divide in to training,testing and validation set randomly. performance, and regression graph is given. How can i defence a question this network is good . Which parameters should i compare to choose best network among many? How should i defense MSE or SSE define the performance of network. To choose my best model performance what parameters should i compare?
Dear Leila,
There are several function to illustrate performance of network such as MSE, SSE and etc .Usually , they have same behavior to define performance of a network. you should choose one of them and compare that for different network. Based on my study, MSE is one of the most usable performance function in articles.
The values of performance and total correlation coefficient help you to check accuracy of your network. Likewise, you can check the network generalization by compare the values of correlation coefficient for train, test and validation data.( the value of them should be near together)
Regards,
Hello Lila,
For selection of best network, most probable criteria is having lowest MSE or SSE. While performing testing of network, you can vary the number of neurons in network and then analyze the MSE value. Then after you can select the network with minimum MSE.
Performing tests or choosing performance model has nothing to do with MATLAB GUI
Perhaps this can be helpful.
There is a chapter of results where several networks were tested. Generalization tests are not available in the MATLAB GUI. Doing it manually is quite difficult, although you can use this idea to propose your own.
http://jupiter.utm.mx/~tesis_dig/12105.pdf
Hi Lila,
1. You must select or make a theoretical model of your network.
2. You must specify and describe the parameters that influence the performance and how they relate.
3. You must determine the values or range of values to be considered by Matlab, and functions that define how these may change.
4. You should generate data through measurement with a protocol analyzer or snifer , or in his absence by the simulation of a test scenario.
5. You now preloading your model and your data, you can use the potential of analysis Matlab to represent and generate a report on network performance.
Hope it help these simple steps.
Hi Lila
A well trained NN has very low mse (the value of performance plot). the best NN has the mse near to zero. Also you can check the regression. it should be more than 0.8.
- Badges
- Science topic