Hi Uma,
To find and visualize the decision boundary for a two-class classficiation problem using Extreme Learning in MATLAB you can follow these steps:
Steps to Implement ELM and Plot Decision Boundary in MATLAB
- Prepare the Dataset: Use a dataset with two features for simplicity. You can use synthetic data or any suitable dataset.
- Implement the ELM: Implement the ELM by initializing the input weights and biases, computing the hidden layer output matrix, and calculating the output weights.
- Generate a Grid of Points: Create a grid of points covering the feature space to evaluate the classifier.
- Predict Class Labels: Use the trained ELM to predict class labels for each point in the grid.
- Plot the Decision Boundary: Visualize the decision boundary using contour plotting.
Below is an example MATLAB script to demonstrate these steps:
% Generate synthetic data
rng(1); % For reproducibility
n = 100;
X = [randn(n, 2) + 1; randn(n, 2) - 1];
y = [ones(n, 1); -ones(n, 1)];
% Standardize features
X = (X - mean(X)) ./ std(X);
% ELM Parameters
nHiddenUnits = 10;
% Initialize weights and biases
inputWeights = rand(size(X, 2), nHiddenUnits) * 2 - 1;
biases = rand(1, nHiddenUnits);
% Compute hidden layer output
H = 1 ./ (1 + exp(-(X * inputWeights + biases)));
% Compute output weights
outputWeights = pinv(H) * y;
% Create a grid of points for visualization
[xGrid, yGrid] = meshgrid(linspace(min(X(:,1))-1, max(X(:,1))+1, 100), ...
linspace(min(X(:,2))-1, max(X(:,2))+1, 100));
% Predict class labels for each point in the grid
gridPoints = [xGrid(:), yGrid(:)];
HGrid = 1 ./ (1 + exp(-(gridPoints * inputWeights + biases)));
predictions = HGrid * outputWeights;
% Reshape predictions to match the grid
Z = reshape(predictions, size(xGrid));
% Plot decision boundary and data points
figure;
contourf(xGrid, yGrid, Z, [0, 0], 'LineColor', 'k', 'LineWidth', 1.5);
hold on;
scatter(X(y == 1, 1), X(y == 1, 2), 'ro', 'filled');
scatter(X(y == -1, 1), X(y == -1, 2), 'bo', 'filled');
title('Decision Boundary of ELM');
xlabel('Feature 1');
ylabel('Feature 2');
legend('Decision Boundary', 'Class 1', 'Class -1');
axis tight;
hold off;
Hope this helps!
