I am new to Fuzzy logic. I created membership functions with some rules by using matlab Fuzzy Logic Toolbox. Now i want to train this mamdani fuzzy model Can any body help ?
Create FIS with inputs and outputs. Use around 2/3rd of the data for this.
The first step is done after categorizing inputs into low, medium, high, etc...
Simultaneously construct output member functions matching the input MFS.
Then, build the rules, if two inputs with four gradations each, you need to make 16 rules. Each rule should associate an output mf. It is exactly here that FIS becomes data driven.
TTest for outputs, by selecting only inputs. If the outpit match witj actuals stop training. If not, readjust the spacing and shape of MFS, and test the outputs again.
If sayisfied, try validating the system with remaing 1/3 of the
einput data. If you notice too much of a gap between actual and predicted out put, reshuffle the data, repeat the cycle again.
I see... From you example I thought you already had some membership functions in place. Well fuzzy control is not really my area of expertise, but I presume you can find some good papers on this (e.g. Generating fuzzy rules from training instances for fuzzy
classification systems by Chen et al.) Hope this helps.
Training the model requires several modifications to the ranges and overlappings of membership functions, once you define a logical Fuzzy Rule Base.
The rules should be kept strictly logical, rather than modifying it just to suit the data which may make the rule illogical. This is due to the possibility of noisy datapoints in the dataset.
You may please have a look at this application article.
With regards
Ashu
Article Fuzzy logic approach in mode choice modelling for education ...
did you find any solution? i have the same problem. for example Mr. jayaram said " Create FIS with inputs and outputs. Use around 2/3rd of the data for this". i dont know what he means by use around 2/3 of the data for this. i created the input, output and rules, now i did not use any data. can someone gives a simple guidance with clear picture of how its getting trained?
Hi Wazir Armin, why don't you create a new "Question" and provide as many details as possible in the explanation field? This allows other fuzzy logic researchers to have enough information to help and answer your specific question meaningfully.
If you are like Atif Shah, and have created a Mamdani FIS heuristically, why would you want to train another FIS from the collected data using ANFIS?