心脏病预测模型（基于Python的数据挖据）

作者：Abdullah Alrhmoun

该项目的目标是建立一个模型，该模型可以根据描述疾病的特征组合预测心脏病发生的概率。为了实现这一目标，作者使用了瑞士Cleveland Clinic Foundation收集的数据集。该项目中使用的数据集包含针对心脏病的14个特征。数据集显示不同水平的心脏病存在从1​​到4和0没有疾病。我们有303行人数据，13个连续观察不同的症状。此项目研究了不同的经典机器学习模型，以及它们在疾病风险中的发现。

导入依赖库

---

#导入依赖库
import pandas as pnd
import numpy as np
from sklearn import preprocessing
from sklearn import neighbors, datasets
from sklearn import cross_validation
from sklearn.linear_model import SGDClassifier
from sklearn import svm
import operator
from sklearn.cross_validation import KFold

import itertools
import numpy as np
import matplotlib.pyplot as plt

from sklearn import metrics
from sklearn.metrics import confusion_matrix

from sklearn import tree
import seaborn as sns

from IPython.display import Image

%matplotlib inline

载入数据

# 添加列名
header_row = ['age','sex','chest_pain','blood pressure','serum_cholestoral','fasting_blood_sugar',\
               'electrocardiographic','max_heart_rate','induced_angina','ST_depression','slope','vessels','thal','diagnosis']

# 载入数据
heart = pnd.read_csv('processed.cleveland.data.csv', names=header_row)
heart[:5]

#查看数据维度
heart.shape

(303, 14)   #303行人的数据，13个连续观察不同症状。

数据探索

# 计算统计值
heart.describe()

计算某个特征的人数

names_descr = dict()
categorical_columns = ["sex", "chest_pain", "fasting_blood_sugar", "electrocardiographic", "induced_angina", "slope", "vessels", \
                       "thal", "diagnosis"]
for c in categorical_columns:
    print  (heart.groupby([c])["age"].count())

使用简单的均值插补方法预处理数据，将缺失的数据更改为平均值

for c in heart.columns[:-1]:
    heart[c] = heart[c].apply(lambda x: heart[heart[c]!='?'][c].astype(float).mean() if x == "?" else x)
    heart[c] = heart[c].astype(float)   

心脏病类型发现

set(heart.loc[:, "diagnosis"].values)

（0 ：没有疾病；1,2,3,4 代表不同疾病类型）

计算1,2,3,4 levels 之间的相似性

vecs_1 = heart[heart["diagnosis"] == 1 ].median().values[:-2]
vecs_2 = heart[heart["diagnosis"] == 2 ].median().values[:-2]
vecs_3 = heart[heart["diagnosis"] == 3 ].median().values[:-2]
vecs_4 = heart[heart["diagnosis"] == 4 ].median().values[:-2]

print ("Similarity between type 1 and type 2 is ", np.linalg.norm(vecs_1-vecs_2))
print ("Similarity between type 1 and type 3 is ", np.linalg.norm(vecs_1-vecs_3))
print ("Similarity between type 1 and type 4 is ", np.linalg.norm(vecs_1-vecs_4))
print ("Similarity between type 2 and type 3 is ", np.linalg.norm(vecs_2-vecs_3))
print ("Similarity between type 2 and type 4 is ", np.linalg.norm(vecs_2-vecs_4))
print ("Similarity between type 3 and type 4 is ", np.linalg.norm(vecs_3-vecs_4))

sim = {"(1,2)": np.linalg.norm(vecs_1-vecs_2), \
       "(1,3)": np.linalg.norm(vecs_1-vecs_3),\
       "(1,4)": np.linalg.norm(vecs_1-vecs_4),\
       "(2,3)": np.linalg.norm(vecs_2-vecs_3),\
       "(2,4)": np.linalg.norm(vecs_2-vecs_4),\
       "(3,4)": np.linalg.norm(vecs_3-vecs_4)    
      }

# 根据相近值排序
sorted_sim = sorted(sim.items(), key=operator.itemgetter(1))
sorted_sim

可以分别使用每个特征的值来比较心脏病的类型

heart_d = heart[heart["diagnosis"] >= 1 ]
heart_d[:5]

数据预处理

# if "diagnosis" == 0, 没有疾病
# if "diagnosis" >= 1, 有疾病
heart.loc[:, "diag_int"] = heart.loc[:, "diagnosis"].apply(lambda x: 1 if x >= 1 else 0)

#数据标准化
preprocessing.Normalizer().fit_transform(heart)

#划分数据集
heart_train, heart_test, goal_train, goal_test = cross_validation.train_test_split(heart.loc[:,'age':'thal'], \
                                                 heart.loc[:,'diag_int'], test_size=0.33, random_state=0)  

#计算相关系数
corr = heart.corr()
heart.corr()

#绘制热图
cmap = sns.diverging_palette(250, 10, n=3, as_cmap=True)

def magnify():
    return [dict(selector="th",
                 props=[("font-size", "7pt")]),
            dict(selector="td",
                 props=[('padding', "0em 0em")]),
            dict(selector="th:hover",
                 props=[("font-size", "12pt")]),
            dict(selector="tr:hover td:hover",
                 props=[('max-width', '200px'),
                        ('font-size', '12pt')])
]

corr.style.background_gradient(cmap, axis=1)\
    .set_properties(**{'max-width': '80px', 'font-size': '10pt'})\
    .set_caption("Hover to magify")\
    .set_precision(2)\
    .set_table_styles(magnify())

探索可视化

#年龄与血压关系
import matplotlib.pyplot as plt
%matplotlib inline

plt.xlabel("age")
plt.ylabel("blood pressure")

# define title
plt.title("Relationship between age and blood pressure")

# plot
plt.scatter(heart['age'], heart['blood pressure'])
plt.show()

 

建立训练模型并评估LSS参数

# add parameters for grid search
loss = ["hinge", "log"]
penalty = ["l1", "l2"]
alpha = [0.1, 0.05, 0.01]
n_iter = [500, 1000]

# build the models with different parameters and select the best combination for the highest Accuracy
best_score = 0
best_param = (0,0,0,0)
for l in loss:
    for p in penalty:
        for a in alpha:
            for n in n_iter:
                print("Parameters for model", (l,p,a,n))
                lss = SGDClassifier(loss=l, penalty=p, alpha=a, n_iter=n)
                lss.fit(heart_train, goal_train)
                print("Linear regression SGD Cross-Validation scores:")
                scores = cross_validation.cross_val_score(lss, heart.loc[:,'age':'thal'], heart.loc[:,'diag_int'], cv=10)
                print (scores)
                print("Mean Linear regression SGD Cross-Validation score = ", np.mean(scores))
                
                if np.mean(scores) > best_score:
                    best_score = np.mean(scores)
                    best_param = (l,p,a,n)
                    
    
print("The best parameters for model are ", best_param)
print("The Cross-Validation score = ", best_score)

lss_best = SGDClassifier(alpha=0.05, fit_intercept=True, loss='log', n_iter=1000,
penalty='l1')
lss_best.fit(heart_train, goal_train)
print("Linear regression SGD Test score:")
print(lss_best.score(heart_test, goal_test))   

模型验证

# Compute confusion matrix
cnf_matrix = confusion_matrix(goal_test, lss_best.predict(heart_test))
np.set_printoptions(precision=2)

# Plot non-normalized confusion matrix
plt.figure()
plot_confusion_matrix(cnf_matrix, classes=["Heart disease", "No heart disease"],
                      title='Confusion matrix, without normalization')
plt.show()

 

参考：

https://github.com/Sonali1197/Heart-disease-prediction-model