机器学习项目实战：泰坦尼克号乘客生存率预测（一）

项目描述

任务：根据训练集数据中的数据预测泰坦尼克号上哪些乘客能生存下来

数据源：csv文件（train.csv）

数据集预览：

1、英文描述：

2、译文描述：

初步分析

注：代码后紧跟运行结果截图

import pandas as pd
import numpy as np
import matplotlib.pyplot as plt
import seaborn as sns# 导入数据
train = pd.read_csv('train.csv')
print(train.head(10))

# 查看数据基本信息
print(train.info())# 结果：共891条数据，共12列，部分列含有缺失值

# 方便查看哪些特征含缺失值
print(train.isnull().sum())

# 计算不同性别的生存率
survived_sex = train.groupby('Sex')['Survived'].mean()
print(survived_sex)
# 可视化
sns.barplot(x='Sex', y='Survived', data=train, errorbar=None, hue='Sex', palette='viridis')# 结论：女性生存率明显高于男性

# 计算不同年龄段的生存率
train['Age_group'] = pd.cut(train["Age"], bins=[0, 5, 14, 18, 30, 60, 110], labels=['baby', 'child', 'teenager', 'adult', 'old_adult', 'old'], right=False)
survived_age = train.groupby('Age_group', observed=False)['Survived'].mean()
print(survived_age)
# 可视化
order = ['baby', 'child', 'teenager', 'adult', 'old_adult', 'old']
sns.barplot(x='Age_group', y='Survived', data=train, order=order, errorbar=None, hue='Age_group', palette='viridis')
plt.show()# 结论：婴儿生存率明显更高，老人生还率明显更低，随年龄的增长，生存率有下降的趋势

# 计算不同 Plass 的乘客的生存率
survival_pclass = train.groupby('Pclass')['Survived'].mean()
print(survival_pclass)
# 可视化
sns.barplot(x='Pclass', y='Survived', data=train, errorbar=None)
plt.show()# 结论：1、2、3号的生存率由高到低排列

# 计算不同票价的乘客的生存率
train['Fare_group'] = pd.cut(train['Fare'], bins=[0, 5, 15, 25, 50, 100, 200, float('inf')], labels=['very low', 'low', 'normal', 'medium', 'high', 'very high', 'Luxury'], right=False)
print(train.groupby('Fare_group', observed=False)['Survived'].mean())
# 可视化
sns.barplot(x='Fare_group', y='Survived', data=train, errorbar=None, hue='Fare_group', palette='viridis')
plt.show()# 结论：票价越高，生存率越高，高价位生存率在 60% 以上

# 查看训练集中，名字特征有哪些头衔（代表乘客所属阶级）
print(train['Name'].str.extract(' ([A-Za-z]+\.)')[0].unique())
# 查看训练集 Cabin 有哪些取值
print('Cabin 仓号首字母有：', train['Cabin'].unique())

数据预处理与特征工程

# 特征工程from sklearn.impute import KNNImputer
from sklearn.preprocessing import StandardScalerdef pre(df):# 家庭人数df['family'] = df["Parch"] + df["SibSp"]# 提取名字头衔，表示阶级信息df['Title'] = df['Name'].str.extract(' ([A-Za-z]+\.)')df['Title'] = df['Title'].str.replace('.', '')df['Title'] = df['Title'].replace(['Don', 'Lady', 'Capt', 'Col', 'Dr', 'Major', 'Rev', 'Sir', 'Jonkheer', 'Dona'], 'Rare')df['Title'] = df['Title'].replace(['Mlle', 'Ms'], 'Miss')df['Title'] = df['Title'].replace('Mme', 'Mrs')# 缺失值处理df['Embarked'].fillna(df['Embarked'].mode()[0])   # 众数填充df['Fare'].fillna(df['Fare'].median())   # 中位数填充# 手动对 Fare 进行独热编码df['very low'] = (df['Fare']<5)df['low'] = (df['Fare']<15) & (df['Fare']>=5)df['normal'] = (df['Fare']<25) & (df['Fare']>=15)df['medium'] = (df['Fare']<50) & (df['Fare']>=25)df['high'] = (df['Fare']<100) & (df['Fare']>=50)df['very high'] = (df['Fare']<300) & (df['Fare']>=100)df['Luxury'] = (df['Fare']>=300)# 填充 Cabin 缺失值，并用仓号第一个字母替代原来的仓号df['Cabin'] = df['Cabin'].fillna('N').map(lambda x: x[0])# KNN 算法填充 Age 的缺失值imputer = KNNImputer(n_neighbors=5)df['Age'] = imputer.fit_transform(df[['Age']])# 对 Age 分组（加上填补的值进分组里），并手动进行标签编码df['Age_group'] = pd.cut(df["Age"], bins=[0, 5, 14, 18, 30, 60, 110], labels=[1, 2, 3, 4, 5, 6], right=False)# 独热编码df = pd.get_dummies(df, columns=['Embarked', 'Title', 'Cabin'], drop_first=True)df['Sex'] = df['Sex'].map({'male': 0, 'female': 1})# 处理特征出现在测试集而不在训练集的可能情况for p in ['A', 'B', 'C', 'D', 'E', 'F', 'G', 'T']:col = f'Cabin_{p}'if col not in df.columns:   # 若存在训练集没出现过的字母，则新建一列df[col] = pd.Series([False]*df.shape[0])# 选择对目标变量可能有影响的特征（挑选特征）cols = ['Pclass', 'Title_Miss', 'Title_Mr', 'Title_Mrs','Title_Rare','Sex','Age', 'Age_group','SibSp', 'Parch','Fare', 'very low', 'low', 'normal', 'medium', 'high', 'very high', 'Luxury','Cabin_A', 'Cabin_B', 'Cabin_C', 'Cabin_D', 'Cabin_E', 'Cabin_F','Cabin_G', 'Cabin_T'  # 忽略缺失值 N ]df2 = df[cols].copy()# 标准化数值特征num_features = ['Age', 'Fare', 'Parch', 'SibSp', 'Age_group']scaler = StandardScaler()df2[num_features] = scaler.fit_transform(df2[num_features])return df2

建模过程

# 建模、模型评估（简单建模与评估）from sklearn.model_selection import train_test_split
from sklearn.linear_model import LogisticRegression
from sklearn.svm import SVC
from sklearn.neighbors import KNeighborsClassifier
from sklearn.ensemble import RandomForestClassifier, GradientBoostingClassifier
from sklearn.metrics import accuracy_scoretrain = pd.read_csv('train.csv')
# 特征工程，调用前面定义的函数
train_ = pre(train)   
X = train_  #特征变量
y = train['Survived']   #目标变量
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.2)models = {'逻辑回归': LogisticRegression(),'支持向量机SVM': SVC(),'KNN': KNeighborsClassifier(),'随机森林': RandomForestClassifier(),'梯度提升树': GradientBoostingClassifier()
}
accuary_cores = {'逻辑回归': 0,'支持向量机SVM': 0,'KNN': 0,'随机森林': 0,'梯度提升树': 0
}# 计算模型在各自默认参数下反复训练 n 次的平均准确率
n = 20
for i in range(n):for model_name, model in models.items():model.fit(X_train, y_train)y_pred = model.predict(X_test)accuracy = accuracy_score(y_test, y_pred)accuary_cores[model_name] += accuracy
print('默认参数下各模型的平均准确率：')
for model_name, accuracy in accuary_cores.items():accuracy /= nprint(f'{model_name}:  {accuracy:.4f}')