点击 “AladdinEdu，同学们用得起的【H卡】算力平台”，H卡级别算力，按量计费，灵活弹性，顶级配置，学生专属优惠。

决策树/SVM/KNN算法对比 × 模型评估指标解析
读者收获：掌握经典机器学习全流程

当80%的机器学习问题可用Scikit-learn解决，掌握其核心流程将成为你的核心竞争力。本文通过对比实验揭示算法本质，带你一站式打通机器学习任督二脉。

一、Scikit-learn全景图：3大核心模块解析

在这里插入图片描述

1.1 算法选择矩阵

在这里插入图片描述

1.2 环境极速配置

# 创建专用环境  
conda create -n sklearn_env python=3.10  
conda activate sklearn_env  # 安装核心库  
pip install numpy pandas matplotlib seaborn scikit-learn  # 验证安装  
import sklearn  
print(f"Scikit-learn version: {sklearn.__version__}")

二、分类实战：鸢尾花识别

2.1 数据探索与预处理

from sklearn.datasets import load_iris  
import pandas as pd  # 加载数据集  
iris = load_iris()  
df = pd.DataFrame(iris.data, columns=iris.feature_names)  
df['target'] = iris.target  # 数据概览  
print(f"样本数: {df.shape[0]}")  
print(f"特征数: {df.shape[1]-1}")  
print(f"类别分布:\n{df['target'].value_counts()}")  # 可视化分析  
import seaborn as sns  
sns.pairplot(df, hue='target', palette='viridis')

2.2 三大分类器对比实验

from sklearn.model_selection import train_test_split  
from sklearn.tree import DecisionTreeClassifier  
from sklearn.svm import SVC  
from sklearn.neighbors import KNeighborsClassifier  # 划分数据集  
X = df.drop(columns='target')  
y = df['target']  
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.2, random_state=42)  # 初始化分类器  
models = {  "决策树": DecisionTreeClassifier(max_depth=3),  "SVM": SVC(kernel='rbf', probability=True),  "KNN": KNeighborsClassifier(n_neighbors=5)  
}  # 训练与评估  
results = {}  
for name, model in models.items():  model.fit(X_train, y_train)  y_pred = model.predict(X_test)  results[name] = y_pred

2.3 分类结果可视化

import matplotlib.pyplot as plt  
from sklearn.metrics import confusion_matrix  # 绘制混淆矩阵  
fig, axes = plt.subplots(1, 3, figsize=(18, 5))  
for i, (name, y_pred) in enumerate(results.items()):  cm = confusion_matrix(y_test, y_pred)  sns.heatmap(cm, annot=True, fmt='d', cmap='Blues', ax=axes[i])  axes[i].set_title(f"{name} 混淆矩阵")  
plt.show()

三、回归实战：波士顿房价预测

3.1 数据解析与特征工程

from sklearn.datasets import fetch_openml  # 加载数据集  
boston = fetch_openml(name='boston', version=1)  
df = pd.DataFrame(boston.data, columns=boston.feature_names)  
df['PRICE'] = boston.target  # 关键特征分析  
corr = df.corr()['PRICE'].sort_values(ascending=False)  
print(f"与房价相关性最高的特征:\n{corr.head(5)}")  # 特征工程  
df['RM_LSTAT'] = df['RM'] / df['LSTAT']  # 创造新特征

3.2 回归模型对比

from sklearn.linear_model import LinearRegression  
from sklearn.tree import DecisionTreeRegressor  
from sklearn.svm import SVR  # 划分数据集  
X = df.drop(columns='PRICE')  
y = df['PRICE']  
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.2, random_state=42)  # 初始化回归器  
regressors = {  "线性回归": LinearRegression(),  "决策树回归": DecisionTreeRegressor(max_depth=5),  "支持向量回归": SVR(kernel='rbf')  
}  # 训练与预测  
predictions = {}  
for name, reg in regressors.items():  reg.fit(X_train, y_train)  pred = reg.predict(X_test)  predictions[name] = pred

3.3 回归结果可视化

# 绘制预测值与真实值对比  
plt.figure(figsize=(15, 10))  
for i, (name, pred) in enumerate(predictions.items(), 1):  plt.subplot(3, 1, i)  plt.scatter(y_test, pred, alpha=0.7)  plt.plot([y_test.min(), y_test.max()], [y_test.min(), y_test.max()], 'r--')  plt.xlabel('真实价格')  plt.ylabel('预测价格')  plt.title(f'{name} 预测效果')  
plt.tight_layout()

四、模型评估指标深度解析

4.1 分类指标四维分析

在这里插入图片描述
鸢尾花分类评估实例：

from sklearn.metrics import precision_score, recall_score, f1_score, accuracy_score  metrics = []  
for name, y_pred in results.items():  metrics.append({  "模型": name,  "准确率": accuracy_score(y_test, y_pred),  "精确率": precision_score(y_test, y_pred, average='macro'),  "召回率": recall_score(y_test, y_pred, average='macro'),  "F1": f1_score(y_test, y_pred, average='macro')  })  metrics_df = pd.DataFrame(metrics)  
print(metrics_df)

在这里插入图片描述

4.2 回归指标三维对比

波士顿房价评估实例：

from sklearn.metrics import mean_squared_error, mean_absolute_error, r2_score  reg_metrics = []  
for name, pred in predictions.items():  reg_metrics.append({  "模型": name,  "MSE": mean_squared_error(y_test, pred),  "MAE": mean_absolute_error(y_test, pred),  "R²": r2_score(y_test, pred)  })  reg_metrics_df = pd.DataFrame(reg_metrics)  
print(reg_metrics_df)

在这里插入图片描述

五、算法原理对比揭秘

5.1 决策树：可解释性之王

核心参数调优指南：

params = {  'max_depth': [3, 5, 7, None],  'min_samples_split': [2, 5, 10],  'criterion': ['gini', 'entropy']  
}  best_tree = GridSearchCV(  DecisionTreeClassifier(),  param_grid=params,  cv=5,  scoring='f1_macro'  
)  
best_tree.fit(X_train, y_train)

5.2 SVM：高维空间的分割大师

核函数选择策略：
在这里插入图片描述

5.3 KNN：简单高效的惰性学习

距离度量对比：

distance_metrics = [  ('euclidean', '欧氏距离'),  ('manhattan', '曼哈顿距离'),  ('cosine', '余弦相似度')  
]  for metric, name in distance_metrics:  knn = KNeighborsClassifier(n_neighbors=5, metric=metric)  knn.fit(X_train, y_train)  score = knn.score(X_test, y_test)  print(f"{name} 准确率: {score:.4f}")

六、模型优化实战技巧

6.1 特征工程：性能提升关键

波士顿房价特征优化：

from sklearn.preprocessing import PolynomialFeatures  # 创建多项式特征  
poly = PolynomialFeatures(degree=2, include_bias=False)  
X_poly = poly.fit_transform(X)  # 新特征训练  
lr_poly = LinearRegression()  
lr_poly.fit(X_train_poly, y_train)  
r2 = lr_poly.score(X_test_poly, y_test)  
print(f"R²提升: {reg_metrics_df.loc[0,'R²']:.2f} → {r2:.2f}")

6.2 交叉验证：防止过拟合

from sklearn.model_selection import cross_val_score  # 5折交叉验证  
scores = cross_val_score(  SVC(),  X, y,  cv=5,  scoring='accuracy'  
)  
print(f"平均准确率: {scores.mean():.4f} (±{scores.std():.4f})")

6.3 网格搜索：自动化调参

from sklearn.model_selection import GridSearchCV  # 定义参数网格  
param_grid = {  'C': [0.1, 1, 10, 100],  'gamma': [1, 0.1, 0.01, 0.001],  'kernel': ['rbf', 'linear']  
}  # 执行搜索  
grid = GridSearchCV(SVC(), param_grid, refit=True, verbose=3)  
grid.fit(X_train, y_train)  
print(f"最优参数: {grid.best_params_}")

七、工业级部署方案

7.1 模型持久化

import joblib  # 保存模型  
joblib.dump(best_model, 'iris_classifier.pkl')  # 加载模型  
clf = joblib.load('iris_classifier.pkl')  # 在线预测  
new_data = [[5.1, 3.5, 1.4, 0.2]]  
prediction = clf.predict(new_data)  
print(f"预测类别: {iris.target_names[prediction[0]]}")

7.2 构建预测API

from flask import Flask, request, jsonify  app = Flask(__name__)  
model = joblib.load('iris_classifier.pkl')  @app.route('/predict', methods=['POST'])  
def predict():  data = request.get_json()  features = [data['sepal_length'], data['sepal_width'],  data['petal_length'], data['petal_width']]  prediction = model.predict([features])  return jsonify({'class': iris.target_names[prediction[0]]})  if __name__ == '__main__':  app.run(host='0.0.0.0', port=5000)

7.3 性能监控仪表盘

from sklearn.metrics import plot_roc_curve, plot_precision_recall_curve  # 分类性能可视化  
fig, ax = plt.subplots(1, 2, figsize=(15, 6))  
plot_roc_curve(model, X_test, y_test, ax=ax[0])  
plot_precision_recall_curve(model, X_test, y_test, ax=ax[1])

八、避坑指南：常见错误解决方案

8.1 数据预处理陷阱

问题：测试集出现未知类别
解决方案：

from sklearn.preprocessing import OneHotEncoder  # 训练阶段  
encoder = OneHotEncoder(handle_unknown='ignore')  
encoder.fit(X_train_categorical)  # 测试阶段自动忽略未知类别  
X_test_encoded = encoder.transform(X_test_categorical)

8.2 特征尺度问题

症状：SVM/KNN性能异常
处方：

from sklearn.preprocessing import StandardScaler  scaler = StandardScaler()  
X_train_scaled = scaler.fit_transform(X_train)  
X_test_scaled = scaler.transform(X_test)  # 注意：只变换不拟合

8.3 样本不均衡处理

解决方案对比：
在这里插入图片描述

结语：机器学习工程师的成长之路

当你在Scikit-learn中完整实现从数据加载到模型部署的全流程，已超越70%的入门者。但真正的进阶之路刚刚开始。

下一步行动指南：

# 1. 复现经典论文算法  
from sklearn.linear_model import LogisticRegression  
model = LogisticRegression(penalty='l1', solver='liblinear')  # 2. 参加Kaggle竞赛  
from kaggle import api  
api.competitions_list(search='getting started')  # 3. 构建个人项目组合  
projects = [  {"name": "鸢尾花分类器", "type": "分类", "accuracy": 0.97},  {"name": "房价预测", "type": "回归", "R2": 0.85}  
]

记住：在机器学习领域，理论认知的深度=代码实践的厚度。现在运行你的第一个完整流程，让Scikit-learn成为你AI旅程中最可靠的伙伴。

附录：Scikit-learn速查表

任务类型	导入路径	核心参数
分类	`from sklearn.ensemble import RandomForestClassifier`	`n_estimators, max_depth`
回归	`from sklearn.linear_model import LinearRegression`	`fit_intercept, normalize`
聚类	`from sklearn.cluster import KMeans`	`n_clusters, init`
降维	`from sklearn.decomposition import PCA`	`n_components`
模型选择	`from sklearn.model_selection import GridSearchCV`	`param_grid, cv`
数据预处理	`from sklearn.preprocessing import StandardScaler`	`with_mean, with_std`