当前位置：首页 > article >正文

MATLAB实战：机器学习分类回归示例

article 2025/8/2 18:14:07

以下是一个使用MATLAB的Statistics and Machine Learning Toolbox实现分类和回归任务的完整示例代码。代码包含鸢尾花分类、手写数字分类和汽车数据回归任务，并评估模型性能。

%% 加载内置数据集
% 鸢尾花数据集（分类）
load fisheriris;
X_iris = meas; % 150x4 特征矩阵
Y_iris = species; % 150x1 类别标签

% 手写数字数据集（分类）
digitDatasetPath = fullfile(matlabroot, 'toolbox', 'nnet', 'nndemos', ...
'nndatasets', 'DigitDataset');
imds = imageDatastore(digitDatasetPath, ...
'IncludeSubfolders', true, 'LabelSource', 'foldernames');
[trainImgs, testImgs] = splitEachLabel(imds, 0.7, 'randomized');

% 提取HOG特征
numTrain = numel(trainImgs.Files);
hogFeatures = zeros(numTrain, 324); % HOG特征维度
for i = 1:numTrain
img = readimage(trainImgs, i);
hogFeatures(i, :) = extractHOGFeatures(img);
end
trainLabels = trainImgs.Labels;

% 汽车数据集（回归）
load carsmall;
X_car = [Weight, Horsepower, Cylinders]; % 100x3 特征矩阵
Y_car = MPG; % 100x1 响应变量

%% 鸢尾花分类任务
rng(1); % 设置随机种子保证可重复性
cv = cvpartition(Y_iris, 'HoldOut', 0.3);
idxTrain = training(cv);
idxTest = test(cv);

% 训练KNN模型
knnModel = fitcknn(X_iris(idxTrain,:), Y_iris(idxTrain), 'NumNeighbors', 5);
knnPred = predict(knnModel, X_iris(idxTest,:));
knnAcc = sum(strcmp(knnPred, Y_iris(idxTest))) / numel(idxTest)

% 训练决策树
treeModel = fitctree(X_iris(idxTrain,:), Y_iris(idxTrain));
treePred = predict(treeModel, X_iris(idxTest,:));
treeAcc = sum(strcmp(treePred, Y_iris(idxTest))) / numel(idxTest)

% 训练SVM
svmModel = fitcecoc(X_iris(idxTrain,:), Y_iris(idxTrain));
svmPred = predict(svmModel, X_iris(idxTest,:));
svmAcc = sum(strcmp(svmPred, Y_iris(idxTest))) / numel(idxTest)

% 混淆矩阵可视化
figure;
confusionchart(Y_iris(idxTest), knnPred, 'Title', 'KNN Confusion Matrix');

%% 手写数字分类（使用KNN示例）
% 训练KNN模型
knnDigitModel = fitcknn(hogFeatures, trainLabels, 'NumNeighbors', 3);

% 处理测试集
numTest = numel(testImgs.Files);
testFeatures = zeros(numTest, 324);
testLabels = testImgs.Labels;
for i = 1:numTest
img = readimage(testImgs, i);
testFeatures(i, :) = extractHOGFeatures(img);
end

% 预测并评估
digitPred = predict(knnDigitModel, testFeatures);
digitAcc = sum(digitPred == testLabels) / numel(testLabels)

%% 回归任务（汽车数据）
rng(2);
cv_car = cvpartition(length(Y_car), 'HoldOut', 0.25);
idxTrain_car = training(cv_car);
idxTest_car = test(cv_car);

% 线性回归
lmModel = fitlm(X_car(idxTrain_car,:), Y_car(idxTrain_car));
lmPred = predict(lmModel, X_car(idxTest_car,:));
lmMSE = loss(lmModel, X_car(idxTest_car,:), Y_car(idxTest_car))

% 多项式回归（二次项）
polyModel = fitlm(X_car(idxTrain_car,:), Y_car(idxTrain_car), 'poly2');
polyPred = predict(polyModel, X_car(idxTest_car,:));
polyMSE = loss(polyModel, X_car(idxTest_car,:), Y_car(idxTest_car))

% 可视化回归结果
figure;
scatter(Y_car(idxTest_car), lmPred, 'b');
hold on;
scatter(Y_car(idxTest_car), polyPred, 'r');
plot([0,50], [0,50], 'k--');
xlabel('Actual MPG');
ylabel('Predicted MPG');
legend('Linear', 'Polynomial', 'Ideal');
title('Regression Results Comparison');

关键函数说明：

分类模型训练:
- fitcknn(): K近邻分类器
- fitctree(): 决策树分类器
- fitcecoc(): 多类SVM分类器
回归模型训练:
- fitlm(): 线性/多项式回归
- 'poly2'参数: 指定二次多项式项
评估指标:
- confusionchart(): 可视化混淆矩阵
- loss(): 计算均方误差（回归）
- 准确率 = 正确预测数/总样本数（分类）