二、代码实现
# /usr/bin/python
# -*- encoding:utf-8 -*-
# data analysis
import pandas as pd
import numpy as np
import random as rnd
# visualization
import seaborn as sns
import matplotlib.pyplot as plt
# machine learning
from sklearn.linear_model import LogisticRegression
from sklearn.svm import SVC, LinearSVC
from sklearn.ensemble import RandomForestClassifier
from sklearn.neighbors import KNeighborsClassifier
from sklearn.naive_bayes import GaussianNB
from sklearn.linear_model import Perceptron
from sklearn.linear_model import SGDClassifier
from sklearn.tree import DecisionTreeClassifier
from sklearn.ensemble import RandomForestRegressor
train_df = pd.read_csv("E:\\pyworkpace\\CDA\\data\\tele_cust_train.csv")
test_df = pd.read_csv("E:\\pyworkpace\\CDA\\data\\tele_cust_test.csv")
combine = [train_df, test_df]
# 查看数据的完整性
train_df.info()
test_df.info()
# 查看特征值的分布
train_df.describe()
train_df.describe(include=['O'])
# 地区与对应流失率
train_df[['area', 'churn']].groupby(['area'], as_index=False).mean().sort_values(by='churn', ascending=False)
# 离散变量和连续变量画图
# 居住面积、租房类型和婚姻状况的关系
grid = sns.FacetGrid(train_df, row='marital', size=2.2, aspect=1.6)
grid.map(sns.pointplot, 'dwllsize', 'churn', 'dwlltype', palette='deep')
grid.add_legend()
plt.show()
# 离散变量补充
# （1）eqpdays 以众数补齐,删除错误值，例如为负数的值,清空再补（112条为负的数据）
freq_port = train_df.eqpdays.dropna().mode()[0] # 众数
for dataset in combine:
dataset['eqpdays'] = dataset['eqpdays'].replace([-5, -4, -3, -2, -1], np.nan)
dataset['eqpdays'] = dataset['eqpdays'].fillna(freq_port)
# （2）dulband 以众数补齐数据
freq_port = train_df.dualband.dropna().mode()[0] # 众数
for dataset in combine:
dataset['dualband'] = dataset['dualband'].fillna(freq_port)
# (4)creditcd、truck、ethnic、marital、kid0_2至kid16_16共9个数据同时缺失，这几个变量同时用未知分类代替并生成一个新的变量new，1为补的缺失值，0为有值的数据
for dataset in combine:
dataset['new'] = 0 # 没有填补过的设置为0
dataset.loc[np.where(np.isnan(dataset['truck']))[0], 'new'] = 1
dataset['truck'] = dataset['truck'].replace(np.nan, 3)
dataset['ethnic'] = dataset['ethnic'].replace(np.nan, 'Z')
dataset['marital'] = dataset['marital'].fillna('Z')
dataset['creditcd'] = dataset['creditcd'].fillna('Z')
dataset[['kid0_2', 'kid3_5', 'kid6_10', 'kid11_15', 'kid16_17']] = \
dataset[['kid0_2', 'kid3_5', 'kid6_10', 'kid11_15', 'kid16_17']].fillna('Y') # 将为空的替换为Y（1）
# 将U替换成0，Y替换成1
dataset['kid0_2'] = dataset['kid0_2'].map({'U': 0, 'Y': 1}).astype(int)
dataset['kid3_5'] = dataset['kid3_5'].map({'U': 0, 'Y': 1}).astype(int)
dataset['kid6_10'] = dataset['kid6_10'].map({'U': 0, 'Y': 1}).astype(int)
dataset['kid11_15'] = dataset['kid11_15'].map({'U': 0, 'Y': 1}).astype(int)
dataset['kid16_17'] = dataset['kid16_17'].map({'U': 0, 'Y': 1}).astype(int)
# 生成新的kids（孩子个数）用于取代kid0_2至kid16_16
for dataset in combine:
dataset['kids'] = dataset.apply(lambda x: x['kid0_2'] + x['kid3_5'] + x['kid6_10'] + x['kid11_15'] + x['kid16_17'],
axis=1)
# 连续变量补充
new_df = train_df.append(test_df, sort=True)
new_df = new_df.reset_index(drop=True) # 重排索引
corrmat = new_df.corr() # 得到连续变量间的相关关系
f, ax = plt.subplots(figsize=(20, 9))
sns.heatmap(corrmat, vmax=0.8, square=True) # 绘制关系矩阵图
'''
补充da_Mean的数据 （其他连续变量以此类似）
与此相关关系较大的变量有：'adjmou','adjrev','avgrev','avgmou','totcalls','avg3mou','avg3rev','totmou'，peak_vce_Mean，mou_Mean，rev_Mean
'''
# 补充da_Mean数据
train_df_da_Mean = new_df.dropna(subset=['da_Mean']).copy()
test_df_da_Mean = new_df[np.isnan(new_df['da_Mean'])]
X_train = train_df_da_Mean[['adjmou','adjrev','avgrev','avgmou','totcalls','avg3mou','avg3rev','totmou']]
Y_train = train_df_da_Mean["da_Mean"]
X_test = test_df_da_Mean[['adjmou','adjrev','avgrev','avgmou','totcalls','avg3mou','avg3rev','totmou']]
model = RandomForestRegressor(n_estimators=100, oob_score=True, criterion='mse')
model.fit(X_train, Y_train.ravel())
Y_test= model.predict(X_test)
r2 = model.score(X_train, Y_train)
test_df_da_Mean['da_Mean'] = Y_test
neww = test_df_da_Mean.append(train_df_da_Mean) # 未改变索引值
neww = neww.sort_index()
new_df['da_Mean'] = neww['da_Mean']
# 编码
new_df['dualband'] = new_df['dualband'].map({'Y':0, 'N':1, 'T':2, 'U':3}).astype(int)
new_df['creditcd'] = new_df['creditcd'].map({'Y':0, 'N':1, 'Z':3}).astype(int)
new_df['ethnic'] = new_df['ethnic'].map({'U':0, 'N':1, 'H':2, 'Z':3, 'F':4, 'S':5, 'R':6, 'O':7, 'G':8, 'J':9,
'P':10, 'I':11, 'B':12, 'D':13, 'X':14, 'C':15, 'M':16}).astype(int)
new_df['marital'] = new_df['marital'].map({'B':0,'M':1,'U':2,'A':3,'S':4,'Z':5}).astype(int)
new_df = new_df.drop(['Customer_ID', 'HHstatin', 'area','kid0_2', 'kid3_5', 'kid6_10', 'kid11_15', 'kid16_17'],axis=1)
# 没有处理'numbcars','income','dwlltype','dwllsize','adult','ownrent','hnd_price'这几个变量
yy = new_df.copy()
yy = yy.drop(['numbcars','income','dwlltype','dwllsize','adult','ownrent','hnd_price'],axis=1)
train_df = yy.dropna(subset=['churn']).copy()
test_df = yy[np.isnan(new_df['churn'])]
X_train=train_df.drop(['churn'],axis=1)
Y_train=train_df['churn']
X_test = test_df.drop(['churn'],axis=1)
# 用决策树
decision_tree = DecisionTreeClassifier()
decision_tree.fit(X_train, Y_train)
Y_pred = decision_tree.predict(X_test)
acc_decision_tree = round(decision_tree.score(X_train, Y_train) * 100, 2)