1. %% I. 清空环境变量
    2. clear all
    3. clc
    4. warning off
    6. %% II. 导入数据
    7. load data.mat
    9. %%
    10. % 1. 随机产生训练集/测试集
    11. a = randperm(569);
    12. Train = data(a(1:500),:);
    13. Test = data(a(501:end),:);
    15. %%
    16. % 2. 训练数据
    17. P_train = Train(:,3:end);
    18. T_train = Train(:,2);
    20. %%
    21. % 3. 测试数据
    22. P_test = Test(:,3:end);
    23. T_test = Test(:,2);
    25. %% III. 创建决策树分类器
    26. ctree = ClassificationTree.fit(P_train,T_train);
    28. %%
    29. % 1. 查看决策树视图
    30. view(ctree);
    31. view(ctree,'mode','graph');
    33. %% IV. 仿真测试
    34. T_sim = predict(ctree,P_test);
    36. %% V. 结果分析
    37. count_B = length(find(T_train == 1));
    38. count_M = length(find(T_train == 2));
    39. rate_B = count_B / 500;
    40. rate_M = count_M / 500;
    41. total_B = length(find(data(:,2) == 1));
    42. total_M = length(find(data(:,2) == 2));
    43. number_B = length(find(T_test == 1));
    44. number_M = length(find(T_test == 2));
    45. number_B_sim = length(find(T_sim == 1 & T_test == 1));
    46. number_M_sim = length(find(T_sim == 2 & T_test == 2));
    47. disp(['病例总数：' num2str(569)...
    48.       '  良性：' num2str(total_B)...
    49.       '  恶性：' num2str(total_M)]);
    50. disp(['训练集病例总数：' num2str(500)...
    51.       '  良性：' num2str(count_B)...
    52.       '  恶性：' num2str(count_M)]);
    53. disp(['测试集病例总数：' num2str(69)...
    54.       '  良性：' num2str(number_B)...
    55.       '  恶性：' num2str(number_M)]);
    56. disp(['良性乳腺肿瘤确诊：' num2str(number_B_sim)...
    57.       '  误诊：' num2str(number_B - number_B_sim)...
    58.       '  确诊率p1=' num2str(number_B_sim/number_B*100) '%']);
    59. disp(['恶性乳腺肿瘤确诊：' num2str(number_M_sim)...
    60.       '  误诊：' num2str(number_M - number_M_sim)...
    61.       '  确诊率p2=' num2str(number_M_sim/number_M*100) '%']);
    63. %% VI. 叶子节点含有的最小样本数对决策树性能的影响
    64. leafs = logspace(1,2,10);
    66. N = numel(leafs);
    68. err = zeros(N,1);
    69. for n = 1:N
    70.     t = ClassificationTree.fit(P_train,T_train,'crossval','on','minleaf',leafs(n));
    71.     err(n) = kfoldLoss(t);
    72. end
    73. plot(leafs,err);
    74. xlabel('叶子节点含有的最小样本数');
    75. ylabel('交叉验证误差');
    76. title('叶子节点含有的最小样本数对决策树性能的影响')
    78. %% VII. 设置minleaf为13，产生优化决策树
    79. OptimalTree = ClassificationTree.fit(P_train,T_train,'minleaf',13);
    80. view(OptimalTree,'mode','graph')
    82. %%
    83. % 1. 计算优化后决策树的重采样误差和交叉验证误差
    84. resubOpt = resubLoss(OptimalTree)
    85. lossOpt = kfoldLoss(crossval(OptimalTree))
    87. %%
    88. % 2. 计算优化前决策树的重采样误差和交叉验证误差
    89. resubDefault = resubLoss(ctree)
    90. lossDefault = kfoldLoss(crossval(ctree))
    92. %% VIII. 剪枝
    93. [~,~,~,bestlevel] = cvLoss(ctree,'subtrees','all','treesize','min')
    94. cptree = prune(ctree,'Level',bestlevel);
    95. view(cptree,'mode','graph')
    97. %%
    98. % 1. 计算剪枝后决策树的重采样误差和交叉验证误差
    99. resubPrune = resubLoss(cptree)
    100. lossPrune = kfoldLoss(crossval(cptree))

    data.zip
    原文链接