决策树处理医疗数据

发表于 | 分类于

决策树处理医疗数据

一、目标

将所给医疗数据进行预处理，根据病人的四个属性——性别、年龄、病程及术中责任血管，对其与即刻面抽之间的联系进行分析。利用决策树分类方法编程实现数据的分类并绘图。

二、算法思想

在数据预处理中进行如下处理：

性别(gender)：F为1，M为2。年龄(age)：0-40岁为1，40-50岁为2，50-60岁为3，60-70岁为4。病程(course)：0-3年为1，3-6年为2，6-9年为3，9-12年为4，12年以上为5。术中责任血管(vessel)：AICA为 1，PICA为2，AICA、PICA为3，AICA、VA为4，PICA、CA为5，其余为0。是否即刻面抽：是为1，否为0。

在决策树 ID3 算法中，通过求概率进而求信息熵，通过求信息熵进而求信息增益，以信息增益为标准挑选信息增益最大的因素作为节点。在创建树的过程中进行遍历，如果类别完全相同则停止划分，如果遍历完所有特征值则选取出现次数最多的类标签。最后以字典类型输出数据。

在决策树绘图中，首先计算出树中叶子节点的个数和树的深度，然后对节点和节点间部分的绘制分别进行函数定义。最后在整个决策树的绘制过程中，利用深度和叶子节点个数对图像区域进行划分，以防止出现绘图中节点过密的情况产生。

三、核心代码

1、决策树算法

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
import operator
import math as mh 

def entropy(dataset):
	num = len(dataset)
	label = {}
	for vec in dataset:
		labelC = vec[-1]
		if labelC not in label.keys():
			label[labelC] = 0
		label[labelC] += 1
	ent = 0.0
	for key in label:
		pro = float(label[key]) / num
		ent -= pro * mh.log(pro, 2)
	return ent

def split(dataset, temp, value):
	dataAf = []
	for vec in dataset:
		if vec[temp] == value:
			vecAf = vec[: temp]
			vecAf.extend(vec[temp+1:])
			dataAf.append(vecAf)
	return dataAf

def choose(dataset):
	allent = entropy(dataset)
	numlab = len(dataset[0]) - 1
	gainmax = 0.0
	feature = -1
	for i in range(numlab):
		listAf = [example[i] for example in dataset]
		listU = set(listAf)
		entAf = 0.0
		for value in listU:
			datasetS = split(dataset, i, value)
			pro = len(datasetS) / float(len(dataset))
			entAf += pro * entropy(datasetS)
		gain = allent - entAf
		if (gain > gainmax):
			gainmax = gain
			feature = i 
	return feature

def most(listR):
	listcount = {}
	for vote in listR:
		if vote not in listcount.keys():
			listcount[vote] = 0
		listcount[vote] += 1
	listSort = sorted(listcount.items(), key = operator.itemgetter(1), reverse = True)
	return listSort[0][0]

def create(dataset, label):                                                           
    listR = [example[-1] for example in dataset]
    if listR.count(listR[0]) == len(listR):
	    return listR[0]
    if (len(dataset[0]) == 1):
	    return most(listR)
    featureCH = choose(dataset)
    featureLabel = label[featureCH]
    tree = {featureLabel: {}}
    del(label[featureCH])
    featureValue = [example[featureCH] for example in dataset]
    listU = set(featureValue)
    for value in listU:
        labelfin = label[:]
        tree[featureLabel][value] = create(split(dataset, featureCH, value), labelfin)
    return tree

2、决策树绘制

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
import matplotlib.pyplot as plt

DN = dict(boxstyle="round4", fc="0.8")
LN = dict(boxstyle="round4", fc="0.8")
line = dict(arrowstyle="<-")

def number(tree):
    num = 0
    size1 = list(tree.keys())
    str1 = size1[0]
    dict2 = tree[str1]
    for key in dict2.keys():
        if type(dict2[key]).__name__=='dict':
            num += number(dict2[key])
        else:   
            num +=1
    return num

def depth(tree):
    depthmax = 0
    size1 = list(tree.keys())
    str1 = size1[0]
    dict2 = tree[str1]
    for key in dict2.keys():
        if type(dict2[key]).__name__=='dict':
            depthcnow = 1 + depth(dict2[key])
        else:   
            depthcnow = 1
        if depthcnow > depthmax: 
            depthmax = depthcnow
    return depthmax

def plotnode(name, center, parents, label):
    PLOT.ax1.annotate(name, xy=parents,  xycoords='axes fraction',
             xytext=center, textcoords='axes fraction',
             va="center", ha="center", bbox=label, arrowprops=line )
    
def plotmid(center, parents, name):
    midx = (parents[0]-center[0])/2 + center[0]
    midy = (parents[1]-center[1])/2 + center[1]
    PLOT.ax1.text(midx, midy, name, va="center", ha="center", rotation=30)

def plottree(tree, parents, name):
    num = number(tree) 
    size1 = list(tree.keys())
    str1 = size1[0]  
    center = (plottree.x + (1 + float(num))/2/plottree.num, plottree.y)
    plotnode(str1, center, parents, DN)
    plotmid(center, parents, name)
    dict2 = tree[str1]
    plottree.y = plottree.y - 1/plottree.depth
    for key in dict2.keys():
        if type(dict2[key]).__name__=='dict':  
            plottree(dict2[key],center,str(key))       
        else:  
            plottree.x = plottree.x + 1/plottree.num
            plotnode(dict2[key], (plottree.x, plottree.y), center, LN)
            plotmid((plottree.x, plottree.y), center, str(key))
    plottree.y = plottree.y + 1/plottree.depth

def PLOT(tree):
    axis = dict(xticks=[], yticks=[])
    PLOT.ax1 = plt.subplot(frameon=False, **axis)  
    plottree.num = float(number(tree))
    plottree.depth = float(depth(tree))
    plottree.x = -0.5/plottree.num; 
    plottree.y = 1;
    plottree(tree, (0.5,1), '')
    plt.show()

四、结果图像

  1. 决策树

    决策树

  2. 绘制后决策树

    绘制后决策树