【Kaggle】泰坦尼克号生存预测 Titanic

前言

官网链接： Titanic - Machine Learning from Disaster | Kaggle

资源链接： 【免费】Kaggle泰坦尼克号生存预测Titanic资源-CSDN文库

Notebook 链接： Titanic Analysis Predictions | LR, DT, RF, GBT | Kaggle

（其中 Version 1-3 含有分析过程，文末仅贴有逻辑回归模型的完整 python 代码）

案例背景

泰坦尼克号的沉没是历史上最臭名昭著的沉船事故之一。

1912 年 4 月 15 日，在她的处女航中，被广泛认为“不沉”的泰坦尼克号与冰山相撞后沉没。不幸的是，船上没有足够的救生艇，导致 2224 名乘客和机组人员中有 1502 人死亡。

虽然生存有一定的运气成分，但似乎某些群体比其他群体更有可能生存。

在本次挑战中，我们要求建立一个预测模型来回答以下问题：“什么样的人更有可能生存？”使用乘客数据（即姓名、年龄、性别、社会经济阶层等）。

数据集介绍

数据分为两组：

• 训练集（train.csv）
• 测试集（test.csv）

训练集：包含机上部分乘客（确切地说是 891 名）的详细信息，重要的是，将揭示他们是否幸存，也称为“基本事实”。

测试集：包含类似的信息，但没有披露每位乘客的“基本事实”。预测这些结果是你的工作。

列名	含义
PassengerId	乘客编号
Survived	生存情况（0：死亡，1：存活）
Pclass	客舱等级
Name	姓名
Sex	性别
Age	年龄
SibSp	同代直系亲属数
Parch	不同代直系亲属数
Ticket	船票编号
Fare	船票价格
Cabin	客舱号
Embarked	登船港口

加载数据集

# 忽略警告
import warnings
warnings.filterwarnings("ignore")

import pandas as pd
# 加载数据集
df = pd.read_csv("./titanic/train.csv")
df.sample(5, random_state=0)

探索性数据分析（EDA）

df.info()

可视化特征和目标值之间关系

from matplotlib import pyplot as plt
import seaborn as sns

features = ["Pclass", "Age", "SibSp", "Parch", "Fare"]
fig, axes = plt.subplots(1, 5, figsize=(15, 3), tight_layout=True)
for feature, ax in zip(features, axes):
    plt.sca(ax)
    sns.kdeplot(df.loc[df["Survived"] == 1, feature], label="1", fill=True)
    sns.kdeplot(df.loc[df["Survived"] == 0, feature], label="0", fill=True)
    plt.legend(title="Survived")
plt.show()

缺失值分析

df.isnull().sum()

# 删除缺失值
data = df['Age'].dropna()

# 绘制直方图
sns.histplot(data, kde=True, color='skyblue', label='Histogram', stat='density')

# 绘制正态分布曲线
sns.kdeplot(data, color='r', label='Normal Distribution')
plt.legend()
plt.show()

sum(df['Cabin'].isnull()) / len(df)

plt.pie(x=df['Embarked'].value_counts().values, labels=df['Embarked'].value_counts().index, autopct='%1.1f%%')
plt.show()

• 处理缺失值的策略
  • Age 趋近于正态分布，根据 Name 中的称呼给 Age 赋其对于均值
  • Cabin 中缺失值占比 77%，缺失过多，删除该列
  • Embarked 中有 2 个缺失值使用占比最大的 S 填充

数据预处理

数据清洗

缺失值处理

import re

def name_title(x):
    return x.split('.')[0].split(' ')[-1]

df['Name'].apply(name_title).value_counts()

df['NameTitle'] = df['Name'].apply(name_title)
df.sample(5, random_state=0)

# 根据分组计算平均值
group_means = df.groupby('NameTitle')['Age'].mean()

# 填充缺失值
df['Age'] = df['Age'].fillna(df['NameTitle'].map(group_means))
df.sample(5, random_state=0)
df = df.drop('Cabin', axis=1)
df['Embarked'].fillna('S', inplace=True)
df.head()

# 提取每个单元格中包含的非字母字符
symbols_per_cell = df['Name'].apply(lambda x: ''.join([char for char in x if not char.isalpha()]))

# 获取所有不同的符号
unique_symbols = set(''.join(symbols_per_cell))
unique_symbols

去除噪声并且规范化文本内容

def remove_noise(x):
    return re.sub(r'[".,()]+', '', x)

def ticket_pref(x):
    if len(x.split(' ')) == 1:
        return 'nan'
    else:
        x = ".".join(x.split(' ')[:-1])
        return re.sub(r'[./]+', '', x).lower()

def ticket_ID(x):
    x = x.split(' ')[-1]
    return int(x) if x.isdigit() else 0

df['Name'] = df['Name'].apply(remove_noise)
df['TicketPref'] = df['Ticket'].apply(ticket_pref)
df['TicketID'] = df['Ticket'].apply(ticket_ID)
df.sample(5, random_state=0)
y = df['Survived']
X = df.drop(['PassengerId', 'Survived', 'Ticket', 'NameTitle'], axis=1)
X.sample(5, random_state=0)

数据转换

• 处理文本数据
  • Name 使用 TF-IDF（Term Frequency-Inverse Document Frequency）进行特征提取（Feature Extraction）
  • Sex、Embarked、TicketPref 使用独热编码（One-Hot Encoding）进行特征编码（Feature Encoding）

from sklearn.compose import ColumnTransformer
from sklearn.pipeline import Pipeline
from sklearn.feature_extraction.text import TfidfVectorizer
from sklearn.preprocessing import OneHotEncoder, StandardScaler
from sklearn.impute import SimpleImputer
from sklearn.compose import make_column_selector as selector

# 提取数值类型的特征列
numeric_columns = selector(dtype_include='number')

# 定义 Pipeline 中每个步骤
text_transformer = Pipeline(steps=[
    ('tfidf', TfidfVectorizer())
])

categorical_transformer = Pipeline(steps=[
    ('onehot', OneHotEncoder(handle_unknown='ignore'))
])

numeric_transformer = Pipeline(steps=[
    ('scaler', StandardScaler())
])

# 使用 ColumnTransformer 指定每列的处理方式
preprocessor = ColumnTransformer(
    transformers=[
        ('text', text_transformer, 'Name'),
        ('categorical', categorical_transformer, ['Sex', 'Embarked', 'TicketPref']),
        ('numeric', numeric_transformer, numeric_columns)
    ])

# 创建完整的 Pipeline
pipeline = Pipeline(steps=[('preprocessor', preprocessor)])

# 在你的数据上使用 Pipeline 进行处理
X_processed = pipeline.fit_transform(X)

数据划分

from sklearn.model_selection import train_test_split

# 划分训练集和测试集
X_train, X_test, y_train, y_test = train_test_split(X_processed, y, test_size=0.2, random_state=0)
X_train.shape, X_test.shape, y_train.shape, y_test.shape

建模

逻辑回归模型

from sklearn.linear_model import LogisticRegression
from sklearn.model_selection import GridSearchCV
from sklearn.metrics import make_scorer, accuracy_score, classification_report
import numpy as np

# 创建逻辑回归模型
lr = LogisticRegression()

# 定义参数网格
param_grid = {
    'C': np.logspace(-3, 3, 7),
    'max_iter': list(range(5, 40, 5)),
}

# 设置多类分类评估器
scorer = make_scorer(accuracy_score)

# 创建 GridSearchCV 对象
grid_search = GridSearchCV(
    estimator=lr,
    param_grid=param_grid,
    scoring=scorer,
    cv=5 # 使用交叉验证
)

# 运行网格搜索
grid_search.fit(X_train, y_train)

# 输出最佳参数
print("Best Parameters: ", grid_search.best_params_)

# 在验证集上评估模型
lr_model = grid_search.best_estimator_
y_pred = lr_model.predict(X_test)

# 评估（Evaluation）
accuracy = accuracy_score(y_test, y_pred)
print(f"Accuracy: {accuracy}")
print(classification_report(y_test, y_pred))

决策分类树模型

from sklearn.tree import DecisionTreeClassifier

# Create Decision Tree classifier
dt_classifier = DecisionTreeClassifier()

# Define parameter grid
param_grid = {
    'criterion': ['gini', 'entropy'],
    'max_depth': list(range(5, 25, 5)),
    'min_samples_split': [3, 7, 12],
    'min_samples_leaf': [2, 4, 6],
}

# Set the scoring metric
scorer = make_scorer(accuracy_score)

# Create GridSearchCV object
grid_search = GridSearchCV(
    estimator=dt_classifier,
    param_grid=param_grid,
    scoring=scorer,
    cv=5  # Using 5-fold cross-validation
)

# Run grid search
grid_search.fit(X_train, y_train)

# Output the best parameters
print("Best Parameters: ", grid_search.best_params_)

# Evaluate the model on the test set
dt_model = grid_search.best_estimator_
y_pred = dt_model.predict(X_test)

# Evaluation
accuracy = accuracy_score(y_test, y_pred)
print(f"Accuracy: {accuracy}")
print(classification_report(y_test, y_pred))

随机森林模型

from sklearn.ensemble import RandomForestClassifier

# Create Random Forest classifier
rf_classifier = RandomForestClassifier()

# Define parameter grid
param_grid = {
    'n_estimators': [50, 100, 150],
    'criterion': ['gini', 'entropy'],
    'max_depth': [5, 10, 15],
    'min_samples_split': [3, 7, 12],
    'min_samples_leaf': [2, 4, 6],
}

# Set the scoring metric
scorer = make_scorer(accuracy_score)

# Create GridSearchCV object
grid_search = GridSearchCV(
    estimator=rf_classifier,
    param_grid=param_grid,
    scoring=scorer,
    cv=5  # Using 5-fold cross-validation
)

# Run grid search
grid_search.fit(X_train, y_train)

# Output the best parameters
print("Best Parameters: ", grid_search.best_params_)

# Evaluate the model on the test set
rf_model = grid_search.best_estimator_
y_pred = rf_model.predict(X_test)

# Evaluation
accuracy = accuracy_score(y_test, y_pred)
print(f"Accuracy: {accuracy}")
print(classification_report(y_test, y_pred))

梯度提升树模型

from sklearn.ensemble import GradientBoostingClassifier

# Create Gradient Boosting classifier
gb_classifier = GradientBoostingClassifier()

# Define parameter grid
param_grid = {
    'n_estimators': [50, 100, 150],
    'learning_rate': [0.01, 0.1, 0.2],
    'max_depth': [3, 4, 5],
    'min_samples_split': [3, 7, 12],
    'min_samples_leaf': [2, 4, 6],
}

# Set the scoring metric
scorer = make_scorer(accuracy_score)

# Create GridSearchCV object
grid_search = GridSearchCV(
    estimator=gb_classifier,
    param_grid=param_grid,
    scoring=scorer,
    cv=5  # Using 5-fold cross-validation
)

# Run grid search
grid_search.fit(X_train, y_train)

# Output the best parameters
print("Best Parameters: ", grid_search.best_params_)

# Evaluate the model on the test set
gb_model = grid_search.best_estimator_
y_pred = gb_model.predict(X_test)

# Evaluation
accuracy = accuracy_score(y_test, y_pred)
print(f"Accuracy: {accuracy}")
print(classification_report(y_test, y_pred))

from sklearn.metrics import roc_curve, roc_auc_score
from sklearn.metrics import confusion_matrix

# 假设 y_test 是真实标签，y_scores 是预测的概率得分
models = [lr_model, dt_model, rf_model, gb_model]
y_scores = [model.predict_proba(X_test)[:, 1] for model in models]

fig, axes = plt.subplots(2, 4, figsize=(15, 7), tight_layout=True)
fig.suptitle('ROC Curve & Confusion matrix', size=16)
for i in range(4):
    # 计算 ROC 曲线的值
    fpr, tpr, thresholds = roc_curve(y_test, y_scores[i])
    # 计算 AUC（Area Under the Curve）
    auc = roc_auc_score(y_test, y_scores[i])
    plt.sca(axes[0][i])
    plt.plot(fpr, tpr, label=f'AUC = {auc:.2f}')
    plt.plot([0, 1], [0, 1], 'k--', label='Random')
    plt.xlabel('False Positive Rate')
    plt.ylabel('True Positive Rate')
    plt.title(models[i].__class__.__name__)
    plt.legend()

    plt.sca(axes[1][i])
    y_pred = models[i].predict(X_test)
    cm = confusion_matrix(y_test, y_pred)
    sns.heatmap(cm, annot=True, cmap='Blues', fmt='g', linewidths=.5)
    plt.title(models[i].__class__.__name__)
    plt.xlabel('Predicted Labels')
    plt.ylabel('Real Labels')
plt.show()

预测

# 导入数据集
test_data = pd.read_csv("./titanic/test.csv")

# 数据预处理
test_data['NameTitle'] = test_data['Name'].apply(name_title)
group_means = test_data.groupby('NameTitle')['Age'].mean()
test_data['Age'].fillna(df['NameTitle'].map(group_means), inplace=True)
test_data['Fare'].fillna(test_data['Fare'].mean(), inplace=True)
test_data['Name'] = test_data['Name'].apply(remove_noise)
test_data['TicketPref'] = test_data['Ticket'].apply(ticket_pref)
test_data['TicketID'] = test_data['Ticket'].apply(ticket_ID)
test = test_data.drop(['PassengerId', 'Ticket', 'Cabin', 'NameTitle'], axis=1)
test.sample(5, random_state=0)

# 数据转化
X_test_processed = pipeline.transform(test)
X_test_processed.shape

# 模型预测
val = lr_model.predict(X_test_processed)
sub = pd.read_csv("./titanic/gender_submission.csv")
sub['Survived'] = val
sub.to_csv('./titanic/submission.csv', index=False)
print("Your submission was successfully saved!")

LR 完整的 python 代码

import pandas as pd
from sklearn.compose import ColumnTransformer
from sklearn.pipeline import Pipeline
from sklearn.feature_extraction.text import TfidfVectorizer
from sklearn.preprocessing import OneHotEncoder, StandardScaler
from sklearn.compose import make_column_selector as selector
from sklearn.model_selection import train_test_split
from sklearn.linear_model import LogisticRegression
from sklearn.model_selection import GridSearchCV
from sklearn.metrics import make_scorer, accuracy_score, classification_report
import numpy as np
import re
import warnings
warnings.filterwarnings("ignore")

'''Setp 1: Load dataset'''
df = pd.read_csv("titanic/train.csv")


'''Setp 2: Data Preprocessing'''
def name_title(x):
    return x.split('.')[0].split(' ')[-1]

def remove_noise(x):
    return re.sub(r'[".,()]+', '', x)

def ticket_pref(x):
    if len(x.split(' ')) == 1:
        return 'nan'
    else:
        x = ".".join(x.split(' ')[:-1])
        return re.sub(r'[./]+', '', x).lower()

def ticket_ID(x):
    x = x.split(' ')[-1]
    return int(x) if x.isdigit() else 0

# data preprocessing
def preprocessing(df):
    df = df.copy()

    # Missing Data Handling
    df['NameTitle'] = df['Name'].apply(name_title)
    
    # Fill in missing values
    df['Age'].fillna(df['NameTitle'].map(df.groupby('NameTitle')['Age'].mean()), inplace=True)
    df['Embarked'].fillna('S', inplace=True)

    # Remove Noise
    df['Name'] = df['Name'].apply(remove_noise)

    # Standardize Text Content
    df['TicketPref'] = df['Ticket'].apply(ticket_pref)
    df['TicketID'] = df['Ticket'].apply(ticket_ID)

    return df

train_df = preprocessing(df)
y = train_df['Survived']
X = train_df.drop(['PassengerId', 'Survived', 'Ticket', 'NameTitle', 'Cabin'], axis=1)


'''Setp 3: Data Transformation'''
# Extracting columns with numerical features
numeric_columns = selector(dtype_include='number')

# Define each step in the pipeline
text_transformer = Pipeline(steps=[
    ('tfidf', TfidfVectorizer())
])

categorical_transformer = Pipeline(steps=[
    ('onehot', OneHotEncoder(handle_unknown='ignore'))
])

numeric_transformer = Pipeline(steps=[
    ('scaler', StandardScaler())
])

# Use ColumnTransformer to specify the processing method for each column
preprocessor = ColumnTransformer(
    transformers=[
        ('text', text_transformer, 'Name'),
        ('categorical', categorical_transformer, ['Sex', 'Embarked', 'TicketPref']),
        ('numeric', numeric_transformer, numeric_columns)
    ])

# Create a complete pipeline
pipeline = Pipeline(steps=[('preprocessor', preprocessor)])

# Use a pipeline to process data
X_processed = pipeline.fit_transform(X)


'''Setp 4: Data Splitting'''
# Splitting the training set and test set
X_train, X_test, y_train, y_test = train_test_split(X_processed, y, test_size=0.2, random_state=0)


'''Setp 5: Modeling'''
# Create Logistic Regression 
lr = LogisticRegression()

# Define parameter grid
param_grid = {
    'C': np.logspace(-3, 3, 7),
    'max_iter': list(range(5, 50, 1)),
}

# Set the scoring metric
scorer = make_scorer(accuracy_score)

# Create GridSearchCV object
grid_search = GridSearchCV(
    estimator=lr,
    param_grid=param_grid,
    scoring=scorer,
    cv=5 # Using 5-fold cross-validation
)

# Run grid search
grid_search.fit(X_train, y_train)

# Output the best parameters
print("Best Parameters: ", grid_search.best_params_)

# Evaluate the model on the test set
lr_model = grid_search.best_estimator_
y_pred = lr_model.predict(X_test)

# Evaluation
accuracy = accuracy_score(y_test, y_pred)
print(f"Accuracy: {accuracy}")
print(classification_report(y_test, y_pred))


'''Setp 6: Predicting''' 
test_data = pd.read_csv("titanic/test.csv")
test_data.head()

def preprocess_data(df):
    df = df.copy()

    # Missing Data Handling
    df['NameTitle'] = df['Name'].apply(name_title)

    # Fill in missing values
    df['Age'].fillna(train_df['NameTitle'].map(train_df.groupby('NameTitle')['Age'].mean()), inplace=True)
    df['Fare'].fillna(df['Fare'].mean(), inplace=True)

    # Remove Noise
    df['Name'] = df['Name'].apply(remove_noise)

    # Standardize Text Content
    df['TicketPref'] = df['Ticket'].apply(ticket_pref)
    df['TicketID'] = df['Ticket'].apply(ticket_ID)
    
    df = df.drop(['PassengerId', 'Ticket', 'NameTitle', 'Cabin'], axis=1)
    return df

# Data preprocessing
test = preprocess_data(test_data)

# Data Transformation
X_test_processed = pipeline.transform(test)

# Predicting
val = lr_model.predict(X_test_processed)
sub = pd.read_csv("titanic/gender_submission.csv")
sub['Survived'] = val
sub.to_csv('submission.csv', index=False)
print("Your submission was successfully saved!")