决策树算法matlab

对于分类和回归两种 离散属性和连续属性的处理 缺失值的处理

function [tree_model] = buildDecisionTree（train_datatrain_labelfeat_typefeature_indexprediction_type）
%prediction_type means the object type classification or regression
%0 means classification and 1 means regression
num_of_feat = length（feat_type）;
num_of_data = size（train_data1）;

if （num_of_feat == 1） || （length（unique（train_label）） == 1） || num_of_data < 5
    %type域表示节点类型，0表示叶节点，1表示非叶结点
    tree_model.type = 0;
    if prediction_type == 0
        %设置叶节点的预测值
        tree_model.pre = FindMostFeat（train_label）;
    else
        tree_model.pre = mean（train_label）;
    end
    %set the other fileds not be used
    tree_model.split_feat = -1;
    tree_model.split_value = -1;
    tree_model.child = [];
else
    [split_feat split_value] = FindSplitFeat（train_data train_label feat_type0）;
    tree_model.split_feat = feature_index（split_feat）;
    tree_model.type = 1;
    tree_model.pre = FindMostFeat（train_label）;
    if feat_type（split_feat） == 0
        %numeric feature
        tree_model.split_value = split_value;
        train_set1 = train_data（train_data（:split_feat） < split_value:）;
        train_set2 = train_data（train_data（:split_feat） >= split_value:）;
        train_set1（:split_feat） = [];
        train_set2（:split_feat） = [];
        if isempty（train_set1） || isempty（train_set2）
            tree_model.type = 0;
            if prediction_type == 0
            %设置叶节点的预测值
                tree_model.pre = FindMostFeat（train_label）;
            else
                tree_model.pre = mean（train_label）;
            end
            %set the other fileds not be used
            tree_model.split_feat = -1;
            tree_model.split_value = -1;
            tree_model.child = [];
            return
        end
        train_label1 = train_label（train_data（:split_feat） < split_value:）;
        train_label2 = train_label（train_data（:split_feat） >= split_value:）;
        feat_type（split_feat） = [];
        feature_index（split_feat） = [];
        tree_model.child = [buildDecisionTree（train_set1train_label1feat_typefeature_indexprediction_type）buildDecisionTree（train_set2train_label2feat_typefeature_indexprediction_type）];
    else
        %nominal feature
        current_feat_value = split_value;
        num_of_feat_value = length（current_feat_value）;
        tree_model.split_value = split_value;
        tree_model.child = [];
        feat_type（split_feat） = [];
        feature_index（split_feat） = [];
        tree_child_empty = 0;
        for i = 1:num_of_feat_value
            sub_train_set = train_data（train_data（:split_feat） == current_feat_value（i）:）;
            sub_train_set（:split_feat） = [];
            if isempty（sub_train_set）
                tree_child_empty = 1;
            end
        end
        if tree_child_empty
            tree_model.type = 0;
            if prediction_type == 0
            %设置叶节点的预测值
                tree_model.pre = Fi

 属性            大小     日期    时间   名称
----------- ---------  ---------- -----  ----
     文件        1405  2013-10-16 16:58  DecisionTreePrediction.m
     文件        3802  2013-10-16 16:57  buildDecisionTree.m
     文件         371  2013-10-05 13:29  CalcEntropy.m
     文件         259  2013-10-06 21:46  CalcLeaves.m
     文件        1004  2013-10-16 14:18  DecisionTree.m
     文件         391  2013-10-06 11:33  FindMostFeat.m
     文件        4866  2013-10-06 20:57  FindSplitFeat.m
     文件         532  2013-10-05 14:31  choose_split_for_numeric.m
     文件        1038  2013-10-16 14:30  main.m