I am using ANFIS GUI in matlab to predict surface roughness values. I have 5 inputs and 1 output with each containing 246 train data sets and 43 test data sets. I am getting less than 60% accuracy when I compare my trained model against test values. I am also getting few negative values even though I have normalised data and number of parameters are less than input values. Please help.
U can directly use nftool or nntool from Matlab instead of ANFIS. 264+43 is more than sufficient data to predict the surface roughness values by neural network toolbox. In ANFIS, u required to define the rules, depending upon how accurately define the rules on which the accuracy is depend. But in case of nftool GUI u have to directly give the input and output values only. No rules
What is the ratio positive/negative? Maybe they are not balanced.
Other thing that is happening is overfitting. You can try add a validation set.
Another thing you can try is to add more features to your dataset.
- Badges
- Science topic