how to implement fuzzy-KNN code to fault detection

Hi everybody!

please, I come to you once again to seek clarification.

indeed, I want to use the fuzzy-KNN algorithm to detect in a system. I have five categories of defects to detect, namely [No, O, S, PS, SD] corresponding to the values [1 2 3 4 5]

I wrote my algorithm and on this basis I tried a code that I find not correct,

I request your help to solve my problem please. if you can reread and bring your point of view, I will be happy

attached my code and my data.

thanks in advance

Answer 1

[Shubham]

Hi Merlin,

It seems you are implementing a fuzzy-kNN algorithm in MATLAB for fault detection. Here are a few points to help improve your code structure and parameter handling:

• Membership Matrix Initialization: Ensure the "M_y" matrix for the output membership is correctly initialized before the main computation loop to prevent unexpected behavior.
• Distance Calculation and Weighing: When transforming distances, make sure the inverse power applied to "Dist" aligns with the fuzzy parameter "m". This step is crucial for determining the influence of nearby points:

Dist = Dist.^(-1/(m-1));

• Cumulative Sum Calculations: Use "cumsum" instead of "cusum" for cumulative sum calculations:

CuSum_Dist = cumsum(Dist, 1);
CuSum_Mult = cumsum(TR_L .* repmat(Dist, [1, class_n]), 1);

• Class Prediction: YEnsure your "predict_class" variable accurately holds the most likely class. The function for determining the maximum membership class, such as "likelihood2class", should correctly extract the highest membership value for each test sample.
• Reading the Dataset: Use "readmatrix" or "readtable" to load your data from "Dataset.xlsx":

dataset = readmatrix('Dataset.xlsx');

These adjustments should help resolve the primary issues with fuzzy-kNN implementation.

For more information on "cumsum", refer to the following documentation link:

https://mathworks.com/help/matlab/ref/double.cumsum.html

Hope this helps