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:
Hope this helps
