基于决策树的泰坦尼克号幸存者预测

知识点

1.决策树模型可以用于分类预测

实验目的

1.学习建立决策树分类预测模型，比较决策树和逻辑回归模型的差异

2.学习使用LabelEncoder对离散变量进行编码和简单的缺失值处理方法

实验环境

1.Oracle Linux 7.4

2.Python 3

实验数据

---

---

使用pandas读取文件

import pandas as pd
df = pd.read_csv('/root/experiment/datas/titanic.csv',index_col=0)
df.head()

---

---

查看数据的统计描述

df.describe()

---

---

查看缺失值

df.isnull().sum()

---

---

查看数据分布

import matplotlib.pyplot as plt
df.hist(figsize=(10,8))
plt.show()

---

---

去掉无关字段

df = df.drop(['Name','Ticket','Cabin'],axis=1)
df.columns

---

---

填充Age缺失值

mean_age = df['Age'].mean()
tmp = df['Age'].copy()
tmp[df.Age.isnull()] = mean_age
df['Age_fill'] = tmp
del tmp
df = df.drop(['Age'],axis=1)
df.columns

---

---

使用LabelEncoder将离散变量转换为编码

from sklearn.preprocessing import LabelEncoder
le = LabelEncoder()
df_sex = le.fit(df['Sex'])
df['Sex'] = df_sex.transform(df['Sex'])
df = df.dropna()
df_embarked = le.fit(df['Embarked'])
df['Embarked'] = df_embarked.transform(df['Embarked'])
df.info()

---

---

划分自变量和因变量

X = df.loc[:,df.columns!='Survived']
y = df.loc[:,df.columns=='Survived']

划分训练集和测试集

from sklearn.model_selection import train_test_split
X_tr,X_ts,y_tr,y_ts = train_test_split(X,y)
X_tr.shape,X_ts.shape

---

---

建立决策树模型

from sklearn.tree import DecisionTreeClassifier
dtc = DecisionTreeClassifier()
dtc = dtc.fit(X_tr,y_tr)

建立逻辑回归模型

from sklearn.linear_model import LogisticRegression
log = LogisticRegression(class_weight='balanced')
log.fit(X_tr,y_tr.values.ravel())

---

---

对测试集预测，查看混淆矩阵

y_dct_pred = dtc.predict(X_ts)
y_log_pred = log.predict(X_ts)

查看决策树模型混淆矩阵

from sklearn.metrics import classification_report,confusion_matrix
confusion_matrix(y_ts,y_dct_pred)

查看逻辑回归模型混淆矩阵

confusion_matrix(y_ts,y_log_pred)

---

---

打印决策树模型混淆矩阵评分

print(classification_report(y_ts,y_dct_pred))

打印逻辑回归模型混淆矩阵评分

print(classification_report(y_ts,y_log_pred))

---

---

使用5折交叉验证计算决策树模型预测准确率

from sklearn.model_selection import cross_val_score
scores = cross_val_score(dtc,X,y,cv=5,scoring='accuracy')
scores.mean()

使用5折交叉验证计算逻辑回归模型预测准确率 

scores = cross_val_score(log,X,y.values.ravel(),cv=5,scoring='accuracy')
scores.mean()

---

---

.实验结论

1.本试验中，决策树模型在测试集上f1-score得分0.72

2.本试验中，逻辑回归模型在测试集上f1-score得分0.75，略好于决策树模型。

3.5折交叉验证显示，决策树模型准确率0.775。

4.5折交叉验证显示，逻辑回归模型准确率0.775。

5.综上，本试验中，逻辑回归模型和决策树模型差异不大。