My Knn Algorithm Code and while loop is not working. Please help ???

Hello,

I want to ask something about knn algorithm in matlab. I'm trying to use kNN Classification and ı want to find best k with matlab. I did some little loop but it's just working with small data.

My data is called as "Adult Data Set". I take this data on the internet. I have 36179 objects as training data and I have 9044 objects as test data.

I want to make prediction for test data and find the best k value according to the error.

Here is my code:

If you can help me, I will very glad. It's not give any error but also not give any output.

training = xlsread('AdultDataTraining.xlsx','a2:n36179');
trainingclass =xlsread('AdultDataTraining.xlsx','o2:o36179');
test = xlsread('AdultDataTest.xlsx','a2:n9045');
testclass = xlsread('AdultDataTest.xlsx','o2:o9045');
num = 9044;
list_prediction = [];
class_list = [testclass];
error_list = [];
i = 1; 
while i <= 36178
    dissimilarity = 0;
    knn = fitcknn(training,trainingclass,'NumNeighbors',i,'Standardize', 1);
    prediction_test = predict(knn, test(:,:));
    list_prediction = [list_prediction, prediction_test];
    k = 1;
    while k <= length(class_list)
        if class_list(k) == list_prediction(k)
            num = 0;
            dissimilarity = dissimilarity + num; 
        else
            num = 1;
            dissimilarity = dissimilarity + num;
        end
        k = k + 1;
    end
    error = dissimilarity / num;
    error_list = [error_list, error];
    i = i + 1;
end
min(error_list)

[Image Analyst]

You forgot to attach your workbook files with the paper clip icon. I'll check back later for them.

Answer 1

[Anay]

Hi Eda,

I understand that you are not getting any output when you are trying to use KNN algorithm to run predictions on your data. Since you have not provided your data files, I cannot tell if there is any issue with your data. But in the code that you have provided, there seems to be an issue which can be a likely cause for this issue.

The variable “num”, which represents the number of samples in test data set, is being reset in your code unintentionally. This causes wrong calculation of the error and also poses the risk for a divide by zero to occur:

    while k <= length(class_list)
        if class_list(k) == list_prediction(k)
            num = 0;%value of num is getting reset
            dissimilarity = dissimilarity + num; 
        else
            num = 1;%value of num is getting reset
            dissimilarity = dissimilarity + num;
        end
        k = k + 1;
    end
    error = dissimilarity / num;%risk of divide by zero

Also, you are iterating from K = 1 to K = number of training samples which in your case is 36178, where K is the number of nearest neighbours. Having value of K that big for KNN is computationally expensive and requires substantial memory for training and prediction. You can consider having a smaller value for the maximum K for your data, like K = 20.

I hope this helps to solve the issue!