使用Python进行App用户细分

App用户细分是根据用户与App的互动方式对用户进行分组的任务。它有助于找到保留用户，找到营销活动的用户群，并解决许多其他需要基于相似特征搜索用户的业务问题。这篇文章中，将带你完成使用Python进行机器学习的App用户细分任务。

App用户细分

在App用户细分的问题中，我们需要根据用户与App的互动方式对用户进行分组。因此，为了解决这个问题，我们需要根据用户如何使用App来获得有关用户的数据。

导入必要的Python库和数据集：

import plotly.graph_objects as go
import plotly.express as px
import plotly.io as pio
import pandas as pd
pio.templates.default = "plotly_white"data = pd.read_csv("userbehaviour.csv")
print(data.head())

输出

   userid  Average Screen Time  Average Spent on App (INR)  Left Review  \
0    1001                 17.0                       634.0            1   
1    1002                  0.0                        54.0            0   
2    1003                 37.0                       207.0            0   
3    1004                 32.0                       445.0            1   
4    1005                 45.0                       427.0            1   Ratings  New Password Request  Last Visited Minutes       Status  
0        9                     7                  2990    Installed  
1        4                     8                 24008  Uninstalled  
2        8                     5                   971    Installed  
3        6                     2                   799    Installed  
4        5                     6                  3668    Installed

让我们先来看看所有用户的最高、最低和平均屏幕时间：

print(f'Average Screen Time = {data["Average Screen Time"].mean()}')
print(f'Highest Screen Time = {data["Average Screen Time"].max()}')
print(f'Lowest Screen Time = {data["Average Screen Time"].min()}')

输出

Average Screen Time = 24.39039039039039
Highest Screen Time = 50.0
Lowest Screen Time = 0.0

现在让我们来看看所有用户的最高、最低和平均支出金额：

print(f'Average Spend of the Users = {data["Average Spent on App (INR)"].mean()}')
print(f'Highest Spend of the Users = {data["Average Spent on App (INR)"].max()}')
print(f'Lowest Spend of the Users = {data["Average Spent on App (INR)"].min()}')

输出

Average Spend of the Users = 424.4154154154154
Highest Spend of the Users = 998.0
Lowest Spend of the Users = 0.0

现在我们来看看活跃用户和卸载了APP的用户的消费能力和屏幕时间的关系：

figure = px.scatter(data_frame = data, x="Average Screen Time",y="Average Spent on App (INR)", size="Average Spent on App (INR)", color= "Status",title = "Relationship Between Spending Capacity and Screentime",trendline="ols")
figure.show()

在这里插入图片描述
卸载该App的用户平均每天屏幕时间不到5分钟，平均花费不到100。我们还可以看到平均屏幕时间与仍在使用该App的用户的平均支出之间存在线性关系。

现在我们来看看用户给出的评分和平均屏幕时间之间的关系：

figure = px.scatter(data_frame = data, x="Average Screen Time",y="Ratings", size="Ratings", color= "Status", title = "Relationship Between Ratings and Screentime",trendline="ols")
figure.show()

在这里插入图片描述
所以我们可以看到，卸载该应用的用户给该应用的评分最多为5分。与评分更高的用户相比，他们的屏幕时间非常低。所以，这描述了那些不喜欢花更多时间的用户对App的评价很低，并在某个时候卸载它。

App用户细分–查找保留和丢失的用户

现在，让我们继续进行App用户细分，以找到App保留和永远失去的用户。这里将使用机器学习中的K-means聚类算法来完成这项任务：

clustering_data = data[["Average Screen Time", "Left Review", "Ratings", "Last Visited Minutes", "Average Spent on App (INR)", "New Password Request"]]from sklearn.preprocessing import MinMaxScaler
for i in clustering_data.columns:MinMaxScaler(i)from sklearn.cluster import KMeans
kmeans = KMeans(n_clusters=3)
clusters = kmeans.fit_predict(clustering_data)
data["Segments"] = clustersprint(data.head(10))

输出

   userid  Average Screen Time  Average Spent on App (INR)  Left Review  \
0    1001                 17.0                       634.0            1   
1    1002                  0.0                        54.0            0   
2    1003                 37.0                       207.0            0   
3    1004                 32.0                       445.0            1   
4    1005                 45.0                       427.0            1   
5    1006                 28.0                       599.0            0   
6    1007                 49.0                       887.0            1   
7    1008                  8.0                        31.0            0   
8    1009                 28.0                       741.0            1   
9    1010                 28.0                       524.0            1   Ratings  New Password Request  Last Visited Minutes       Status  Segments  
0        9                     7                  2990    Installed         0  
1        4                     8                 24008  Uninstalled         2  
2        8                     5                   971    Installed         0  
3        6                     2                   799    Installed         0  
4        5                     6                  3668    Installed         0  
5        9                     4                  2878    Installed         0  
6        9                     6                  4481    Installed         0  
7        2                     1                  1715    Installed         0  
8        8                     2                   801    Installed         0  
9        8                     4                  4621    Installed         0

现在让我们来看看我们得到的数据划分：

print（data[“Segments”].value_counts（））

输出

0    910
1     45
2     44
Name: Segments, dtype: int64

现在让我们重命名这些数据段，以便更好地理解：

data["Segments"] = data["Segments"].map({0: "Retained", 1: "Churn", 2: "Needs Attention"})

进行数据可视化：

PLOT = go.Figure()
for i in list(data["Segments"].unique()):PLOT.add_trace(go.Scatter(x = data[data["Segments"]== i]['Last Visited Minutes'],y = data[data["Segments"] == i]['Average Spent on App (INR)'],mode = 'markers',marker_size = 6, marker_line_width = 1,name = str(i)))
PLOT.update_traces(hovertemplate='Last Visited Minutes: %{x} <br>Average Spent on App (INR): %{y}')PLOT.update_layout(width = 800, height = 800, autosize = True, showlegend = True,yaxis_title = 'Average Spent on App (INR)',xaxis_title = 'Last Visited Minutes',scene = dict(xaxis=dict(title = 'Last Visited Minutes', titlefont_color = 'black'),yaxis=dict(title = 'Average Spent on App (INR)', titlefont_color = 'black')))