1. Understand and master the logistic regression classification method;

2. Master the effect evaluation of logistic regression model;

3. Master the application scenarios of decision tree classification.

1. This experiment is the analysis and prediction of telecom customer churn rate.

2. Analyze whether users will be lost by analyzing users' packages, calls, traffic, behavior, contracts, associated purchases, service months, etc., and predict the loss possibility of some users,

3. You can launch some new policies or activities for retention according to the situation associated with the user.

1. Logistic regression model and decision tree

Decision tree classification principle decision tree is a process of classifying data through a series of rules. It provides a rule like method of what values will be obtained under what conditions. Decision tree is divided into classification tree and regression tree. Classification tree makes decision tree for discrete variables

2.AUC,fpr,tpr；

AUC (Area Under Curve) is defined as the area enclosed by the coordinate axis under the ROC curve. Obviously, the value of this area will not be greater than 1. Since the ROC curve is generally above the line y=x, the value range of AUC is between 0.5 and 1. The closer the AUC is to 1.0, the higher the authenticity of the detection method; When it is equal to 0.5, the authenticity is the lowest and has no application value.

TPR and FRP are commonly used indicators for classification and detection. TPR is the real rate and FPR is the false positive rate. They are all based on the measurement standard of confusion matrix.

3.ROC curve;

The Chinese name of ROC is receiver operating characteristic curve, which originates from the radar signal analysis technology of World War II.

ROC curve drawing: calculate the FPR and TPR of the model results respectively, and then draw the TPR as the ordinate and TPR as the abscissa to obtain the ROC curve. Each point on the ROC curve corresponds to a threshold

1. Data loading

#1. Data preparation and data loading
import pandas as pd
import numpy as np
import matplotlib.pyplot as plt
plt.rcParams['font.sans-serif'] = ['KaiTi']
plt.rcParams['font.serif'] = ['KaiTi']
data=pd.read_excel("Experiment 3 data.xlsx")
data1=data
data1
data1.set_index('ID', inplace = True)

2. Data analysis

Use describe to handle data exceptions and process data

# 2. Data analysis and data preprocessing
#2.2 data outlier description (filtered box diagram)
da=data1.describe()
q1 =da.loc['25%']
q3 = da.loc['75%']
iqr = q3-q1
mi = q1 - 1.5*iqr
ma = q3 + 1.5*iqr
error = da[(da<mi)|(da>ma)]
# print(error)
print('Error:{}'.format(len(error)))
data_c1 = da[(da>=mi)&(da<=ma)]
# print(data_c1)
data_c1.plot.box(title="Abnormal handling box diagram")
plt.grid(linestyle="--", alpha=0.3)
plt.show()

give the result as follows

1. Data preprocessing

# View total customer churn
churnvalue = data1["Lost users"].value_counts()
labels = data1["Lost users"].value_counts().index
plt.pie(churnvalue,
        labels=["Not lost","Drain"],
        explode=(0.1,0),
        autopct='%.2f%%', 
        shadow=True,)
plt.title("Customer churn ratio",size=24)
plt.show()
# It can be seen from the pie chart that the lost customers account for one fifth of the total customers, and the loss rate is 21.90%

From the customer churn ratio chart, we can see that the customer churn ratio is about 21%, accounting for about 1 / 5 of the total customers. If we can reduce the number of customers, the enterprise will reach another height.

Let's look at the specific factors causing customer churn

import seaborn as sns
sns.set(style="whitegrid")
plt.rcParams['font.sans-serif'] = ['KaiTi']
plt.rcParams['font.serif'] = ['KaiTi']
covariables=["Package amount", "Change behavior", "Associated purchase", "Group user"]
fig,axes=plt.subplots(2,2,figsize=(10,12))
for i, item in enumerate(covariables):
 plt.subplot(2,2,(i+1))
    ax=sns.countplot(x=item,hue="Lost users",data=data1,palette="Set2")
    plt.xlabel(str(item),fontsize=16)
    plt.tick_params(labelsize=14)     # Set axis font size
    plt.title("Impact on customer churn "+ str(item))
    i=i+1
plt.tight_layout()
plt.show()

As can be seen from the chart, this is a factor that has a great impact on customers, especially the package amount

Impact of signing a contract or not on customer churn

# Visualization package
plt.rcParams['font.sans-serif'] = ['KaiTi']
plt.rcParams['font.serif'] = ['KaiTi']
fig = plt.figure(figsize=(12,8))
sns.barplot(x="Service contract",
            y="Lost users", 
            data=data1, 
            palette="Pastel1", )
plt.title("Impact of service contract on customer churn")
plt.xlabel('Service contract',fontsize=16)
plt.ylabel('Customer churn',fontsize=16)
plt.tick_params(labelsize=13)     # Set axis font size
plt.show()

#It can be seen from the histogram that the turnover rate of customers who signed the contract is low, partly because the customers who chose to sign the contract are relatively stable,
#Another reason is that the contract has a long-term constraint rate. When the customer signs the contract, preferential promotion can be carried out for the customer signing the contract.

It can be seen from the histogram that the turnover rate of customers who have signed the contract is low, partly because the customers who choose to sign the contract are relatively stable, and also because the contract has a long-term constraint rate. When the customer signs the contract, preferential promotion can be carried out for the customers who sign the contract.

1. Select features and divide the dataset

2. #Correlation analysis and remove attributes with less correlation
   plt.figure(figsize=(16,8))
   data.corr()['Lost users'].sort_values(ascending = False).plot(kind='bar')
   plt.tick_params(labelsize=14)     # Set axis font size
   plt.xticks(rotation=45)         # Sets the x-axis text direction
   plt.title("correlation analysis ",fontsize=20)
   plt.show()

From the figure, the factors that are less relevant to the loss of customers are additional traffic, behavior change and additional call duration, so we can temporarily ignore these three factors when modeling

Based on several factors with strong correlation, the data set is divided to prepare for the establishment of the following model

  2. Prediction modeling

(1) The decision tree is used to model on the training set.

The confusion matrix and ROC curve are used to evaluate the model in the test set.

We evaluate the model based on fpr, tpr, confusion matrix and roc

As we all know, the closer R^2 is to 1, the better the model is. From the perspective of R^2, the prediction rate of the model is still relatively good. It can also be seen from the confusion matrix that most of the predicted values (i.e. dark parts) are concentrated between 0.5 and 1.5, which shows that the accuracy of the model can indeed reach 98%

The same conclusion can be obtained from the AUC value of roc

(2) Modeling of logistic regression algorithm on training set

The confusion matrix and ROC curve are used to evaluate the model in the test set.

From the prediction test results of the above two models, we can find that in fact, the prediction accuracy of the two models is similar, and the accuracy can reach 98%.

Summary of results

After a series of analysis, we can find that there are many external factors affecting customer churn, but the customer churn rate of signed contracts will be particularly low. Therefore, what we need to do is to increase preferential efforts, develop more signed users and firmly grasp old users. Therefore, gradient stratification of preferential policies may be carried out, and preferential policies for those in dispute may get twice the result with half the effort.

Let's share it here today! No, this code reference is not easy to use. It's not as good as my picture!