# optimizing classification using Neural Net

I am using fuzzy input as my input data to neural net. The input neurons are fuzzy members. The output consists of 2 labels. I am just getting the classification accuracy as 83%, as oppose to the 98% accuracy obtained by the author of proposed paper. Can any one tell me how can i increase the accuracy. My input neurons are 74, output neurons are 2 , hidden neurons are 50. my training data consists of 630 records and testing data of 70 records.

Please help me.

Please help me.

0

## Answers

1,869UnicornTo optimize the Neural Net you should try different values for the learning rate, the momentum, and the hidden layers configurations - you see that the Neural Net is a beast that is hard to tame. Maybe you also want to compare it to different algorithms such as e.g. the SVM (optimize C for the linear kernel, C and gamma for the radial kernel).

Best regards,

Marius

4Contributor ISorry my question might seem silly to you, but i am a student.

I am using a different input variables but the output is same and the procedure is exactly same. Does the result get affected by input variable. As my procedure and output is same, i was expecting same result. I am having 74 input neurons, 50 hidden neurons and 2 output.

Can parameter optimization be applied for optimizing the values of learning rate, momentum and training cycles of neural nets.

Sir please help me by giving few suggestions.

1,869Unicornof course the input variables affect the results - actually the Neural Net is built based on them, and once created produces the output based on the input.

To calculate the mentioned parameters you can use the Optimize Parameters (Grid) operators.

Best regards,

Marius

4Contributor I4Contributor II am using the Neural network to tune the fuzzy members. Each time my data changes i have to again change the values of training cycles and hidden layers to get more accuracy. The same value for training cycles and hidden layers does give the optimal solution.

Also, Does the Neural network learning for two class classification depend on the combination ratio of two class in training set. If the class ratio in training set is uneven then will it degrade the performance.

Please help me out by suggesting.

Thanking you.