SVM classification using LIBSVM, Accuracy Problem

I have datasets belongs to two different classes as given in attached xls file. I converted the data into "train data" set and "test data" set in CSV format

I have still doubt on the way i am giving input to the program.

If i run my attached matlab code, it is always showing 50% accuracy only...

Please let me know whether i am giving input data correctly or not, and how to get good accuracy?

here the test data is a part of train data only...

Answer 1

[Aditya]

Hi HST,

When working with neural networks for classification in MATLAB, it’s crucial to properly format your input data and target labels. For a two-class classification problem, your feature matrix should have samples as columns and features as rows. The target labels should be converted to a one-hot encoded format, where each class is represented by a vector (for example, class 1 as [1; 0] and class 2 as [0; 1]). If you use raw class labels (like 1 and 2) instead of one-hot vectors, the network will not be able to learn the classification task correctly, which often leads to an accuracy of around 50%—equivalent to random guessing.

Below is an example of how you should prepare your data and train a simple neural network in MATLAB:

% Load your CSV data
trainData = readmatrix('train.csv');
testData = readmatrix('test.csv');

% Separate features and labels
X_train = trainData(:, 1:end-1)';  % Features as rows, samples as columns
y_train = trainData(:, end);       % Class labels (1 or 2)
X_test  = testData(:, 1:end-1)';
y_test  = testData(:, end);

% Convert labels to one-hot encoding
T_train = full(ind2vec(y_train')); % Now T_train is 2 x N
T_test = full(ind2vec(y_test'));

% Create and train the neural network
net = patternnet(10);
net = train(net, X_train, T_train);

% Predict on test data
Y_pred = net(X_test);

% Convert network output to class labels
[~, predicted_labels] = max(Y_pred, [], 1);

% Calculate accuracy
accuracy = sum(predicted_labels' == y_test) / numel(y_test);
fprintf('Test Accuracy: %.2f%%\n', accuracy*100);