用 Iris数据做决策树分析

Iris数据的准备

1.直接从sklearn.datasets 加载或转化成文件已备本地使用

代码如下：

from sklearn.datasets import load_iris
import pandas as pd

# 加载数据集
iris = load_iris()
df = pd.DataFrame(data=iris.data, columns=iris.feature_names)

# 将数字标签替换为植物名
df['species'] = [iris.target_names[i] for i in iris.target]  # 新增一列植物名
df['target'] = iris.target
# # 保存为Excel文件（不包含行索引）
df.to_excel("iris_dataset.xlsx", index=False)

execl表格如下所示

2.可以在https://archive.ics.uci.edu/dataset/53/iris下载

过程

• 读取数据
• 确定特征
• 训练决策树模型(按重要性分裂）
• 模型评估
• 可视化决策树
• 生成决策树分析报告

示例

代码如下

import pandas as pd
import numpy as np

# 读取数据
df_train = pd.read_excel('iris_dataset.xlsx')

# print(df_train.keys())
# print(df_train.head())


# 特征工程
def create_features(df):
    #初始化features
    features = pd.DataFrame()
    #简化df中原来的列名（feature）
    features['SepalLength'] = df['sepal length (cm)']
    features['SepalWidth'] = df['sepal width (cm)']
    features['PetalLength'] = df['petal length (cm)']
    features['PetalWidth'] = df['petal width (cm)']
    return features
# 创建特征
features_train = create_features(df_train)
# print("features_train")
# define the X and y
X_train = df_train.drop(['species', 'target'], axis=1)
y_train = df_train.loc[:, 'target']
# print(X.head())
# print(y.head())
# print(X_train.shape, y_train.shape)
#
# 训练决策树模型(按重要性分裂）
from sklearn.tree import DecisionTreeClassifier,export_text

dt = DecisionTreeClassifier(
        criterion='entropy',
        min_samples_leaf=5,
        splitter='best'  # 确保优先选重要性高的特征
        # ,max_features=1  # 每次分裂只考虑1个特征（需配合特征选择使用）
)
dt.fit(X_train, y_train)
feature_importance = pd.DataFrame({
    'feature': features_train.columns,
    'importance': dt.feature_importances_})
# print("特征重要性:\n", feature_importance)

# 模型评估
from sklearn.metrics import accuracy_score

y_predict = dt.predict(X_train)
train_score = accuracy_score(y_train, y_predict)
print(train_score)

# 可视化决策树
import matplotlib.pyplot as plt
from sklearn.tree import plot_tree

# 设置中文显示
plt.rcParams['font.sans-serif'] = ['SimHei']
plt.rcParams['axes.unicode_minus'] = False

fig = plt.figure(figsize=(10, 10))
#print(features_train.columns)
plot_tree(dt, feature_names=features_train.columns,class_names=['setosa', 'versicolor','virginica'],
          filled=True, rounded=True, fontsize=12)
#plt.show()
plt.savefig('decision_tree.png', dpi=300, bbox_inches='tight', pad_inches=0.5)

# 生成决策树规则文本
tree_rules = export_text(dt, feature_names=list(features_train.columns))

# 生成分析报告
with open('决策树分析.md', 'w', encoding='utf-8') as f:
    f.write('# 决策树模型分析报告\n\n')

    # 模型性能
    f.write('## 1. 模型性能\n')
    f.write(f'- 训练集准确率: {train_score:.4f}\n')


    # 特征重要性
    f.write('## 2. 特征重要性\n')
    feature_importance = pd.DataFrame({
        'feature': features_train.columns,
        'importance': dt.feature_importances_
    }).sort_values('importance', ascending=False)

    for _, row in feature_importance.iterrows():
        f.write(f'- {row["feature"]}: {row["importance"]:.4f}\n')

    # 决策规则
    f.write('\n## 3. 决策规则\n')
    f.write('```\n')
    f.write(tree_rules)
    f.write('\n```\n\n')

生成的决策树如下：

生成的分析报告如下：

决策树模型分析报告

1. 模型性能

• 训练集准确率: 0.9733

2. 特征重要性

• PetalLength: 0.6777
• PetalWidth: 0.3071
• SepalLength: 0.0151
• SepalWidth: 0.0000

3. 决策规则

|--- PetalLength <= 2.45
|   |--- class: 0
|--- PetalLength >  2.45
|   |--- PetalWidth <= 1.75
|   |   |--- PetalLength <= 4.95
|   |   |   |--- SepalLength <= 5.15
|   |   |   |   |--- class: 1
|   |   |   |--- SepalLength >  5.15
|   |   |   |   |--- class: 1
|   |   |--- PetalLength >  4.95
|   |   |   |--- class: 2
|   |--- PetalWidth >  1.75
|   |   |--- PetalLength <= 4.95
|   |   |   |--- class: 2
|   |   |--- PetalLength >  4.95
|   |   |   |--- class: 2