How to adjust the parameters of membership function after the ANFIS is trained

I use the 'Neuro-Fuzzy Designer' (anfis) to train a dataset of 3 inputs and one output. I have positive inputs and outputs values, but the resulted training and testing outputs include some negative values. I think this is because the membership function for the input 3 ranges from -0,5 to 0.4828 as shown in figure 1. How could I make this membership function ranging from 0 to 1 in order to not consider negative values?

Answer 1

[Sam Chak]

Hi @sarara

If the ANFIS model has been successfully trained and the performance is satisfactory, except for the aesthetic aspect where the 'negative value' appears in the parameter of the triangular membership function (mf1), you can manually modify the parameter using the straightforward steps outlined below:

fis = mamfis('NumInputs', 3, 'NumInputMFs', 3, 'NumOutputs', 1, 'NumOutputMFs', 8);

%% original parameters of mf1
fis.Inputs(1).MembershipFunctions(1)

ans = 
  fismf with properties:

          Type: "trimf"
    Parameters: [-0.4167 0 0.4167]
          Name: "mf1"

plotmf(fis, 'input', 1), grid on
title('Before modifying parameters of mf1')

%% modified parameters of mf1
fis.Inputs(1).MembershipFunctions(1).Parameters(1) = 0;
fis.Inputs(1).MembershipFunctions(1)

ans = 
  fismf with properties:

          Type: "trimf"
    Parameters: [0 0 0.4167]
          Name: "mf1"

figure
plotmf(fis, 'input', 1), grid on
title('After modifying parameters of mf1')