【数据挖掘实战：特征发现与特征提取技术详解】

数据挖掘实战：特征发现与特征提取技术详解

一、技术概念解析

1.1 特征发现（Feature Discovery）

特征发现是探索性数据分析的核心环节，通过以下方法识别潜在有效特征：

• 统计分布分析：偏度、峰度、标准差
• 数据可视化：箱线图、热力图、散点矩阵
• 领域知识驱动：医疗数据中的BMI指数、电商中的购买频次
• 关联规则挖掘：Apriori算法发现特征组合

1.2 特征提取（Feature Extraction）

将原始数据转换为机器学习友好格式的关键步骤：

• 数值处理：标准化、归一化、非线性变换
• 类别编码：One-Hot、Target Encoding、Embedding
• 维度压缩：PCA、t-SNE、Autoencoder
• 时序特征：滑动窗口统计、傅里叶变换
• 文本向量化：TF-IDF、Word2Vec、BERT
• 图像特征：HOG、SIFT、CNN特征图

二、结构化数据处理实战

2.1 数据准备

使用开源信用卡欺诈检测数据集：

import pandas as pd
from sklearn.datasets import fetch_openml

data = fetch_openml('creditcard', version=1, as_frame=True)
df = pd.DataFrame(data.data, columns=data.feature_names)
df['Class'] = data.target

2.2 特征发现流程

1. 异常值检测：

Q1 = df['Amount'].quantile(0.25)
Q3 = df['Amount'].quantile(0.75)
df['Amount_Outlier'] = (df['Amount'] > (Q3 + 1.5*(Q3-Q1))).astype(int)

2. 交互特征生成：

df['V1_V2_Interaction'] = df['V1'] * df['V2']
df['Time_Amount_Ratio'] = df['Time'] / (df['Amount'] + 1e-5)

3. 时间序列特征：

from tsfresh import extract_features
time_features = extract_features(df[['Time', 'Amount']], column_id='Time')

2.3 特征提取技术

1. 非线性变换：

df['Amount_log'] = np.log1p(df['Amount'])
df['V3_square'] = df['V3'] ** 2

2. 分箱处理：

from sklearn.preprocessing import KBinsDiscretizer
est = KBinsDiscretizer(n_bins=5, encode='ordinal', strategy='quantile')
df['Amount_bin'] = est.fit_transform(df[['Amount']])

3. 自动特征生成：

from featuretools import dfs
features = dfs(entityset=es, 
              target_entity='transactions',
              agg_primitives=['sum', 'mean', 'count'],
              trans_primitives=['hour', 'is_weekend'])

三、非结构化数据处理

3.1 文本特征提取

from sklearn.feature_extraction.text import TfidfVectorizer
from gensim.models import Word2Vec

# TF-IDF
tfidf = TfidfVectorizer(ngram_range=(1,2), max_features=5000)
X_tfidf = tfidf.fit_transform(texts)

# 词向量
model = Word2Vec(sentences, vector_size=100, window=5, min_count=3)
word_vectors = model.wv

# BERT特征
from transformers import BertTokenizer, BertModel
tokenizer = BertTokenizer.from_pretrained('bert-base-uncased')
model = BertModel.from_pretrained('bert-base-uncased')
inputs = tokenizer(text, return_tensors="pt", padding=True, truncation=True)
outputs = model(**inputs)
embeddings = outputs.last_hidden_state.mean(dim=1)

3.2 图像特征提取

from keras.applications import ResNet50
from keras.preprocessing import image

model = ResNet50(weights='imagenet', include_top=False)
img = image.load_img('image.jpg', target_size=(224, 224))
x = image.img_to_array(img)
x = np.expand_dims(x, axis=0)
features = model.predict(x)

# 可视化特征图
import matplotlib.pyplot as plt
plt.figure(figsize=(12,8))
for i in range(16):
    plt.subplot(4,4,i+1)
    plt.imshow(features[0,:,:,i], cmap='viridis')
    plt.axis('off')

四、特征评估与选择

4.1 特征重要性分析

from sklearn.ensemble import RandomForestClassif