Qualitative Modelling with ANFIS, FIS Optimization, Loading Data

Hi,

I want to use ANFIS for qualitative modelling purposes. I already built the first FIS structure with fuzzy toolbox. However, I want to take this structure and optimize it with ANFIS and to show that my model is a good model. But, my problem is that ANFIS wants me to load data to train it. But, I am not able to load data because it is qualitative modelling and I dont have any numerical data to load. The only numerical data I use is that 1-10 scale showing the quality (good, medium, bad). So, how can I overcome this "loading data" problem?

Thanks in Advance

Kind Regards Fatih

Answer 1

[Arkadiy Turevskiy]

The only way to show that your model is a good model is to validate it against a set of input/output data. What are the inputs to your system? Sounds like you already know what the output is and how to represent it numerically.

See this section of the doc for more details on training.

[Fatih Ozel]

Thanks for your answer.

My inputs are the experts opinions. But they say (for example) if government incentives increase, there will be more electric vehicles in the system. These are if-then rules which I already incorporated to my model by using fuzzy toolbox. So, I can build FIS with the information I have but I cant train it with ANFIS because I dont have any numerical data (except than 1-10 scale I gave)owing to its qualitative nature. I thought about using experts opinions and expressing them with the scale (1-10) for each input, but I could not find a way to present them to Matlab. Do you have any suggestions for this situation?

Thanks in Advance

Kind Regards

Fatih

[Arkadiy Turevskiy]

Your situation looks very similar to the tipper example we have in the intro section of the doc here. In this example, you are designing a fuzzy system that decides how much to tip in a restaraunt based on quality of food and service. The rules are similar to what you describe, i.e., if service is good, then tip is average. I suggest you go through this example to how to fuzzify input variables. That should help you get started.