A Basic AdaBoost Algorithm Implementation (MATLAB Source Code)

Below is a basic MATLAB implementation of the AdaBoost algorithm:

function [H, alpha] = adaboost(X, Y, T)

% Initialize weight vector uniformly for all samples

N = size(X, 1);

W = ones(N, 1) / N;

% Store weak classifiers and their corresponding weights

H = cell(T, 1);

alpha = zeros(T, 1);

for t = 1:T

% Train a weak classifier (decision stump, SVM, etc.) using current sample weights

h = trainClassifier(X, Y, W);

% Calculate the weighted error rate of the weak classifier

epsilon = computeError(h, X, Y, W);

% Compute classifier weight using AdaBoost formula: alpha = 0.5*ln((1-epsilon)/epsilon)

alpha(t) = 0.5 * log((1 - epsilon) / epsilon);

% Update sample weights: increase weights for misclassified samples

W = updateWeights(h, X, Y, W, alpha(t));

% Store the trained weak classifier

H{t} = h;

end

end

function h = trainClassifier(X, Y, W)

% This function trains a base classifier (e.g., decision tree, SVM) using weighted samples

% Implementation typically involves finding the best feature threshold that minimizes weighted error

% Returns the trained weak classifier model

h = trainedClassifier;

end

function epsilon = computeError(h, X, Y, W)

% Calculate weighted classification error by comparing predictions with true labels

% Error is computed as sum of weights for misclassified samples

% Returns the weighted error rate

epsilon = error;

end

function W = updateWeights(h, X, Y, W, alpha)

% Update sample weights using exponential rule: W_i = W_i * exp(±alpha)

% Increase weights for misclassified samples, decrease for correctly classified ones

% Returns updated sample weight vector (normalized to sum to 1)

W = updatedWeights;

end