In [1]:

1 import pandas as pd
2 import numpy as np
3 import [Link] as plt
4 import seaborn as sns
5 import warnings
6 [Link]('ignore')

In [2]:

1 df=pd.read_csv("customer_data.csv")

In [3]:

1 [Link]()

Out[3]:

ID Year_Birth Education Marital_Status Income Kidhome Teenhome Dt_Customer Recenc

0 5524 1957 Graduation Single 58138.0 0 0 04/09/12 5

1 2174 1954 Graduation Single 46344.0 1 1 08/03/14 3

2 4141 1965 Graduation Together 71613.0 0 0 21/08/13 2

3 6182 1984 Graduation Together 26646.0 1 0 10/02/14 2

4 5324 1981 PhD Married 58293.0 1 0 19/01/14 9

5 rows × 29 columns

 

In [4]:

1 [Link]

Out[4]:

Index(['ID', 'Year_Birth', 'Education', 'Marital_Status', 'Income', 'Kidhome',

'Teenhome', 'Dt_Customer', 'Recency', 'MntWines', 'MntFruits',
'MntMeatProducts', 'MntFishProducts', 'MntSweetProducts',
'MntGoldProds', 'NumDealsPurchases', 'NumWebPurchases',
'NumCatalogPurchases', 'NumStorePurchases', 'NumWebVisitsMonth',
'AcceptedCmp3', 'AcceptedCmp4', 'AcceptedCmp5', 'AcceptedCmp1',
'AcceptedCmp2', 'Complain', 'Z_CostContact', 'Z_Revenue', 'Response'],
dtype='object')

EDA
In [5]:

1 #Seperating Out Date,Month & Year of Customer Enrollment

2 df["Date_enroll"]=df["Dt_Customer"].[Link]("/").str[0]
3 df["Month_enroll"]=df["Dt_Customer"].[Link]("/").str[1]
4 df["Year_enroll"]=df["Dt_Customer"].[Link]("/").str[2]

In [6]:

1 [Link]()

Out[6]:

ID Year_Birth Education Marital_Status Income Kidhome Teenhome Dt_Customer Recenc

0 5524 1957 Graduation Single 58138.0 0 0 04/09/12 5

1 2174 1954 Graduation Single 46344.0 1 1 08/03/14 3

2 4141 1965 Graduation Together 71613.0 0 0 21/08/13 2

3 6182 1984 Graduation Together 26646.0 1 0 10/02/14 2

4 5324 1981 PhD Married 58293.0 1 0 19/01/14 9

5 rows × 32 columns

 

In [7]:

1 #Birth Year Values

2 [Link](figsize=(10,5))
3 [Link](df["Year_Birth"],bins=50,color="red",stat="probability")
4 [Link]("Birth Year Counts")
5 [Link]()

[Link] is of people born between 1940-2000.

2 or 3 variables of data are from year 1900 or earlier it can be result of wrong entry or an outlier.
In [8]:

1 #Types Of Education
2 [Link](figsize=(10,6))
3 [Link](df["Education"].value_counts(),autopct="%1.1f%%",labels=[Link](df["Education"])
4 [Link]("Distribution Of Education")
5 [Link](title="Classes", loc="center left", bbox_to_anchor=(1, 0.5))
6 [Link]("equal")
7 [Link]()

[Link] is divided into 5 types:

[Link] ---16.5%
[Link] ---21.7%
[Link] ---9.1%
[Link] ---2.4%
v.2n Cycle ---50.3%
In [29]:

1 #Marital Status Values

2 [Link](figsize=(10,5))
3 [Link](df["Marital_Status"],bins=8,color="burlywood",element="bars")
4 [Link]("Marital Status")
5 [Link]()

[Link] Status is divided into 8 types:

'Single', 'Together', 'Married', 'Divorced', 'Widow', 'Alone','Absurd', 'YOLO'

In [30]:

1 #Education Type & Income

2 [Link](figsize=(10,5))
3 [Link](x=df["Education"],y=df["Income"],hue=df["Income"],color="deeppink")
4 [Link]("Education Type & Income")
5 [Link]()

[Link] of Basic education is less comapred to others.

[Link] Income generated is from Graduation but it can be wrong entry or outlier as well.

In [31]:

1 #Birth Year VS Income

2 [Link](figsize=(10,5))
3 [Link](data=df, x="Year_Birth", y="Income",hue="Income",palette="cool")
4 [Link]("Birth Year VS Income")
5 [Link]()
1..Birth year wise income is distributed equally but some group of people from birth year 1970-1980 are earning
more compared to others.

In [32]:

1 #Distribution Of Kidhome
2 [Link](figsize=(10,6))
3 [Link](df["Kidhome"].value_counts(),autopct="%1.1f%%",labels=[Link](df["Kidhome"]),sha
4 [Link]("Distribution Of Kidhome")
5 [Link](title="Classes", loc="center left", bbox_to_anchor=(1, 0.5))
6 [Link]("equal")
7 [Link]()

1.50.7% of people do not have kids,40.1% people have 1 child whereas 2.1% people have 2childs.
In [33]:

1 #Distribution Of Teenhome
2 [Link](figsize=(10,6))
3 [Link](df["Teenhome"].value_counts(),autopct="%1.1f%%",labels=[Link](df["Teenhome"]),s
4 [Link]("Distribution Of Teenhome")
5 [Link](title="Classes", loc="center left", bbox_to_anchor=(1, 0.5))
6 [Link]("equal")
7 [Link]()

1.51.7% of people do not have teens at their home,46.0% of people have 1 teen wheras 2.3% of people have 2
teens at home.
In [34]:

1 #Recency
2 [Link](figsize=(19,10))
3 [Link](x=df["Recency"].unique(),y=df["Recency"].value_counts(),color="darkblue")
4 [Link]("Recency frequency")
5 [Link]()

In [35]:

1 #MntWines:Amount Spent On wines

2 #Education & Wines
3 [Link](figsize=(10,5))
4 [Link](data=df,x="Education",y="MntWines",color="blue")
5 [Link]("Education AND Amount spent on wine")
6 [Link]()
In [36]:

1 #Marital Status & Wines

2 [Link](figsize=(10,5))
3 [Link](data=df,x="Marital_Status",y="MntWines",color="red")
4 [Link]("Marital Status AND Amount spent on wine")
5 [Link]()

In [37]:

1 #MntFruits:Amount Spent On Fruits

2 #Education & Fruits
3 [Link](figsize=(10,5))
4 [Link](data=df,x="Education",y="MntFruits",color="blue")
5 [Link]("Education AND Amount spent on fruits")
6 [Link]()

[Link]/Highest amount is spend on wine.

[Link] with basic education are not spending any amount on wines.

In [38]:

1 #Marital Status & fruits

2 [Link](figsize=(10,5))
3 [Link](data=df,x="Marital_Status",y="MntFruits",color="red")
4 [Link]("Marital Status AND Amount spent on fruits")
5 [Link]()

In [39]:

1 #Kidhome & fruits

2 [Link](figsize=(10,5))
3 [Link](data=df,x="Kidhome",y="MntFruits",color="mediumvioletred",ci=None)
4 [Link]("Kidhome AND Amount spent on fruits")
5 [Link]()
In [40]:

1 #Enrollment year wise

2 [Link](figsize=(10,5))
3 [Link](df["Year_enroll"],color="blue")
4 [Link]("Year Of Enrollment")
5 [Link]()

[Link] of the registrations are from year 2013 in dataset provided of 2 years.

In [41]:

1 #Distribution Of MntMeatProducts
2 [Link](figsize=(10,5))
3 [Link](df["MntMeatProducts"].unique(),color="turquoise")
4 [Link]("Distribution Of MntMeatProducts")
5 [Link]()
In [42]:

1 #MntFruits:Amount Spent On MntMeatProducts

2 #Education & MntMeatProducts
3 [Link](figsize=(10,5))
4 [Link](data=df,x="Education",y="MntMeatProducts",color="blue")
5 [Link]("Education AND Amount spent on MntMeatProducts")
6 [Link]()

In [43]:

1 #Marital Status & MntMeatProducts

2 [Link](figsize=(10,5))
3 [Link](data=df,x="Marital_Status",y="MntMeatProducts",color="red")
4 [Link]("Marital Status AND Amount spent on MntMeatProducts")
5 [Link]()
In [44]:

1 # Teenhome & MntMeatProducts

2 [Link](figsize=(10,5))
3 [Link](data=df,x="Teenhome",y="MntMeatProducts",color="green",hue="MntMeatProducts"
4 [Link]("Teenhome AND Amount spent on MntMeatProducts")
5 [Link]()

In [45]:

1 #Distribution Of MntMeatProducts
2 [Link](figsize=(10,5))
3 [Link](df["MntFishProducts"].unique(),color="darkblue")
4 [Link]("Distribution Of MntFishProducts")
5 [Link]()
In [46]:

1 #Distribution Of MntSweetProducts
2 [Link](figsize=(10,5))
3 [Link](df["MntSweetProducts"].unique(),color="red")
4 [Link]("Distribution Of MntSweetProducts")
5 [Link]()

In [47]:

1 # Teenhome & MntSweetProducts

2 [Link](figsize=(10,5))
3 [Link](data=df,x="Teenhome",y="MntSweetProducts",color="red",hue="MntSweetProducts"
4 [Link]("Teenhome AND Amount spent on MntSweetProducts")
5 [Link]()
In [48]:

1 # Kidhome & MntSweetProducts

2 [Link](figsize=(10,5))
3 [Link](data=df,x="Kidhome",y="MntSweetProducts",color="blue",hue="MntSweetProducts"
4 [Link]("Kidhome AND Amount spent on MntSweetProducts")
5 [Link]()

In [49]:

1 #Distribution Of MntGoldProds
2 [Link](figsize=(10,5))
3 [Link](df["MntGoldProds"].unique(),color="gold")
4 [Link]("Distribution Of MntGoldProds")
5 [Link]()
In [50]:

1 # MntGoldProds & Education

2 [Link](figsize=(10,5))
3 [Link](df["Education"],df["MntGoldProds"],color="rebeccapurple")
4 [Link]("Education AND Amount spent on MntGoldProds")
5 [Link]()

In [51]:

1 # Income & MntGoldProds

2 [Link](figsize=(10,5))
3 [Link](df["MntGoldProds"],df["Income"],color="darksalmon",alpha=0.5,edgecolors="blue"
4 [Link]("Amount spend")
5 [Link]("Income")
6 [Link]("Income AND Amount spent on MntGoldProds")
7 [Link]()
In [52]:

1 # Teenhome & MntGoldProds

2 [Link](figsize=(10,5))
3 [Link](df["Teenhome"],df["MntGoldProds"],color="darkmagenta")
4 [Link]( "Teenhome AND Amount spent on MntGoldProds")
5 [Link]()

In [53]:

1 # Kidhome & MntGoldProds

2 [Link](figsize=(10,5))
3 [Link](df["Kidhome"],df["MntGoldProds"],color="chocolate")
4 [Link]( "Kidhome AND Amount spent on MntGoldProds")
5 [Link]()
In [54]:

1 # MntGoldProds & Marital_Status

2 [Link](figsize=(10,5))
3 [Link](df["Marital_Status"],df["MntGoldProds"],color="darkblue")
4 [Link]("Marital_Status AND Amount spent on MntGoldProds")
5 [Link]()

[Link] having 0 teens are spending more on gold followed by 1 and 2.

[Link] having 2 kids are spending less on gold comapred with having 1 kid or none.

3..Single & married are spending more on gold than comapred to others.

In [55]:

1 df["NumDealsPurchases"].values

Out[55]:

array([3, 2, 1, ..., 1, 2, 3], dtype=int64)

In [56]:

1 #NumDealsPurchases
2 [Link](figsize=(10,5))
3 [Link](x=df["NumDealsPurchases"],palette="husl")
4 [Link]("count")
5 [Link]("Distribution of NumDealsPurchases")
6 [Link]()

[Link] Like to purchase when discount is there can be clearly seen that when discount was there many people
have brought the items.
In [57]:

1 [Link](df["Complain"],color="lightslategrey")
2 [Link]("Complaints")
3 [Link]()

[Link] are not much complaints of customer only 21 complaints are there

In [59]:

1 #plotting respones of campaigns

In [134]:

1 color = "cornflowerblue"
2
3 fig, axs = [Link](nrows=2, ncols=2, figsize=(17,10))
4
5 axs[0, 0].hist(df["AcceptedCmp1"], color=color)
6 axs[0, 0].set_title("Campaign 1", fontsize=14)
7 axs[0, 0].set_xlabel("Accepted Campaign 1", fontsize=12)
8 axs[0, 0].set_ylabel("Count", fontsize=12)
9
10 axs[0, 1].hist(df["AcceptedCmp2"], color=color)
11 axs[0, 1].set_title("Campaign 2", fontsize=14)
12 axs[0, 1].set_xlabel("Accepted Campaign 2", fontsize=12)
13 axs[0, 1].set_ylabel("Count", fontsize=12)
14
15 axs[1, 0].hist(df["AcceptedCmp3"], color=color)
16 axs[1, 0].set_title("Campaign 3", fontsize=14)
17 axs[1, 0].set_xlabel("Accepted Campaign 3", fontsize=12)
18 axs[1, 0].set_ylabel("Count", fontsize=12)
19
20 axs[1, 1].hist(df["AcceptedCmp4"], color=color)
21 axs[1, 1].set_title("Campaign 4", fontsize=14)
22 axs[1, 1].set_xlabel("Accepted Campaign 4", fontsize=12)
23 axs[1, 1].set_ylabel("Count", fontsize=12)
24
25 fig.subplots_adjust(hspace=0.4, wspace=0.2)
26 [Link](labels=['Responses'], loc='upper right', fontsize=12)
27 [Link]("Campaign Responses")
28 [Link]()
In [61]:

1 #Plotting response i.e of last campaign

2 [Link](figsize=(10,5))
3 [Link](df["Response"],color="darkblue",element="step")
4 [Link]("Response Of last Campaign")
5 [Link]()

[Link] can be clearly seen people doesnt accept the offer in 2nd campaign ,the ratio is almost same in other 3
campaigns.

[Link] majority of people have not accepted offer in last campaign [Link] are least affected by campaigns.

In [62]:

1 #NumWebPurchases,NumStorePurchases & NumCatalogPurchases

In [63]:

1 #marital Status VS NumWebPurchases

2 [Link](figsize=(10,5))
3 [Link](df["Marital_Status"],df["NumWebPurchases"],color="darkred",edgecolor="black")
4 [Link]("Marital Status")
5 [Link]("NumWebPurchases")
6 [Link]("marital Status VS NumWebPurchases")
7 [Link]()

In [64]:

1 #marital Status VS NumStorePurchases

2 [Link](figsize=(10,5))
3 [Link](df["Marital_Status"],df["NumStorePurchases"],color="slateblue",edgecolor="black")
4 [Link]("Marital Status")
5 [Link]("NumStorePurchases")
6 [Link]("#marital Status VS NumStorePurchases")
7 [Link]()
In [65]:

1 #marital Status VS NumCatalogPurchases

2 [Link](figsize=(10,5))
3 [Link](df["Marital_Status"],df["NumCatalogPurchases"],color="lightskyblue",edgecolor="bl
4 [Link]("Marital Status")
5 [Link]("NumCatalogPurchases")
6 [Link]("marital Status VS NumCatalogPurchases")
7 [Link]()

[Link] purchases are mostly done in majority by single followed by married persons.

[Link] purchases are done equally by people on marital status those are alone,absurd,YOLO are not in large
number anyways.

[Link] purchases are done mostly by married & together.

In [66]:

1 #Education VS NumWebPurchases
2 [Link](figsize=(10,5))
3 [Link](df["Education"],df["NumWebPurchases"],color="lightcoral",edgecolor="black")
4 [Link]("Education")
5 [Link]("NumWebPurchases")
6 [Link]("Education VS NumWebPurchases")
7 [Link]()

In [67]:

1 #Education VS NumStorePurchases
2 [Link](figsize=(10,5))
3 [Link](df["Education"],df["NumStorePurchases"],color="mediumorchid",edgecolor="black")
4 [Link]("Number of Store Purchases")
5 [Link]("Education")
6 [Link]("#Education VS NumStorePurchases")
7 [Link]()
In [68]:

1 #Education VS NumCatalogPurchases
2 [Link](figsize=(10,5))
3 [Link](df["Education"], df["NumCatalogPurchases"], color="olivedrab",edgecolor="black")
4 [Link]("NumCatalogPurchases")
5 [Link]("Education")
6 [Link]("Education VS NumCatalogPurchases")
7 [Link]()

[Link] with basic education are least in catolog purchase.

[Link] with all sort of education visit store but the [Link] people with basic are less.

In [69]:

1 #NumWebVisitsMonth
In [70]:

1 #Education vs NumWebVisitsMonth
2 [Link](figsize=(10,5))
3 [Link](x=df["Education"],y=df["NumWebVisitsMonth"],hue=df["NumWebVisitsMonth"],pale
4 [Link]("Education vs NumWebVisitsMonth")
5 [Link]()

In [71]:

1 #Teenhome vs NumWebVisitsMonth
2
3 [Link](figsize=(10,5))
4 [Link](x=df["Teenhome"],y=df["NumWebVisitsMonth"],hue=df["NumWebVisitsMonth"],palet
5 [Link]("Teenhome vs NumWebVisitsMonth")
6 [Link]()
In [72]:

1 #Marital_Status vs NumWebVisitsMonth
2 [Link](figsize=(15,8))
3
4 [Link](x=df["Marital_Status"],y=df["NumWebVisitsMonth"],hue=df["NumWebVisitsMonth"
5 [Link]("Marital_Status vs NumWebVisitsMonth")
6 [Link]()

In [73]:

1 # Kidhome vs NumWebVisitsMonth
2 [Link](figsize=(10,5))
3
4 [Link](data=df,x="Kidhome",y="NumWebVisitsMonth",hue=df["NumWebVisitsMonth"],palett
5 [Link]("Kidhome vs NumWebVisitsMonth")
6 [Link]()
In [74]:

1 # Income vs NumWebVisitsMonth
2 [Link](figsize=(10,5))
3
4 [Link](data=df,x="Income",y="NumWebVisitsMonth",palette="cool",hue=df["NumWebVisi
5 [Link]("Income vs NumWebVisitsMonth")
6 [Link]()

[Link] sort of people visit website but [Link] people with basic education are less visting comparitively to others.

[Link] trying to find if [Link] visits depends on Teenhome & Kidhome but that was not the case,No sort of relation
found between them.

[Link] Status is also a not a consideration for visiting website people with all sort of marital status visit it,count of
widows and others is less because we have very small [Link] data regarding to them in the dataset.

[Link] with all sort of income are visiting website frequently no pattern identified on one class.
In [75]:

1 #Number of Website Visits vs Number of Website Purchases

2 [Link](figsize=(10,5))
3 [Link](x=df["NumWebVisitsMonth"], y=df["NumWebPurchases"],hue=df["NumWebPurchases
4 [Link]("Number of Website Visits vs Number of Website Purchases")
5 [Link]("Number of Website Visits")
6 [Link]("Number of Website Purchases")
7 [Link]()

Conclusions/Observations:

Birth Year
[Link] is of people born between 1940-2000.

2.2 or 3 variables of data are from year 1900 or earlier it can be result of wrong entry or an outlier.

Education:
[Link] is divided into 5 types:

[Link] ---16.5%
[Link] ---21.7%
[Link] ---9.1%
[Link] ---2.4%
v.2n Cycle ---50.3%

Marital Status:
[Link] Status is divided into 8 types:
 [Link]
[Link]
[Link]
[Link]
[Link]
[Link]
[Link]
iii O O

Income:
[Link] of Basic education is less comapred to others.

[Link] Income generated is from Graduation but it is too high,can be wrong entry or outlier as well.

[Link] year wise income is distributed equally but some group of people from birth year 1970-1980 are earning
more compared to others.

Kids & Teens:

1.50.7% of people do not have kids,40.1% people have 1 child whereas 2.1% people have 2childs.

2.51.7% of people do not have teens at their home,46.0% of people have 1 teen wheras 2.3% of people have 2
teens at home.

Amount Spend:
[Link]/Highest amount is spend on wine.

[Link] with basic education are not spending any amount on wines.

[Link] with basic education are not spending much on fruits & meat as comapared to others

[Link] types of people differentiated on basis of martial status are spending nearby same on fruits& meat products.

Num Deals Purchase (Discount)

[Link] Like to purchase when discount is there can be clearly seen that when discount was there many people
have brought the items.

Registrations:
Most of the registrations are from year 2013 in dataset provided of 2 years.

Complaints:
There are not much complaints of customer only 21 complaints are there

Amount Spent On Gold:

[Link] having 0 teens are spending more on gold followed by 1 and 2.

[Link] having 2 kids are spending less on gold comapred with having 1 kid or none.
3..Single & married are spending more on gold than comapred to others.

Offer Campaigns:
[Link] can be clearly seen people doesnt accept the offer in 2nd campaign ,the ratio is almost same in other 3
campaigns.

[Link] majority of people have not accepted offer in last campaign [Link] are least affected by campaigns.

Purchases:
[Link] purchases are mostly done in majority by single followed by married persons.

[Link] purchases are done equally by people on marital status those are alone,absurd,YOLO are not in large
number anyways.

[Link] purchases are done mostly by married & together.

[Link] with basic education are least in catolog purchase.

Store & Website Visit

[Link] sort of people visit website but [Link] people with basic education are less visting comparitively to others.

[Link] trying to find if [Link] visits depends on Teenhome & Kidhome but that was not the case,No sort of relation
found between them.

[Link] Status is also a not a consideration for visiting website people with all sort of marital status visit it,count of
widows and others is less because we have very small [Link] data regarding to them in the dataset.

[Link] with all sort of education visit store but the [Link] people with basic are less.

[Link] with all sort of income are visiting website frequently no pattern identified on one class.

[Link] of the purchases are done in 1-9 visits.

DATA PREPROCESSING
In [76]:

1 #Dropping Date,Month & Year Columns Seperated for EDA purpose

2 #Dropping ID As well because of high cardinality
3 #Dropping Dt_Customer as its in str format & even its not of that important

In [77]:

1 [Link](["ID","Date_enroll","Month_enroll","Year_enroll","Dt_Customer"],axis=1,inplace=Tru
In [78]:

1 #Checking for duplicates

2 [Link]().sum()

Out[78]:

182

In [79]:

1 #Dropping Duplicates
2 df.drop_duplicates(inplace=True)

In [80]:

1 [Link]().sum()

Out[80]:

In [81]:

1 #All duplicates values have been removed

In [82]:

1 #Checking For null values

2 [Link]().sum()

Out[82]:

Year_Birth 0
Education 0
Marital_Status 0
Income 24
Kidhome 0
Teenhome 0
Recency 0
MntWines 0
MntFruits 0
MntMeatProducts 0
MntFishProducts 0
MntSweetProducts 0
MntGoldProds 0
NumDealsPurchases 0
NumWebPurchases 0
NumCatalogPurchases 0
NumStorePurchases 0
NumWebVisitsMonth 0
AcceptedCmp3 0
AcceptedCmp4 0
AcceptedCmp5 0
AcceptedCmp1 0
AcceptedCmp2 0
Complain 0
Z_CostContact 0
Z_Revenue 0
Response 0
dtype: int64
In [83]:

1 #As Income have 24 null values we are going to drop them

2 [Link](inplace=True)

In [84]:

1 [Link]().sum()

Out[84]:

Year_Birth 0
Education 0
Marital_Status 0
Income 0
Kidhome 0
Teenhome 0
Recency 0
MntWines 0
MntFruits 0
MntMeatProducts 0
MntFishProducts 0
MntSweetProducts 0
MntGoldProds 0
NumDealsPurchases 0
NumWebPurchases 0
NumCatalogPurchases 0
NumStorePurchases 0
NumWebVisitsMonth 0
AcceptedCmp3 0
AcceptedCmp4 0
AcceptedCmp5 0
AcceptedCmp1 0
AcceptedCmp2 0
Complain 0
Z_CostContact 0
Z_Revenue 0
Response 0
dtype: int64
In [85]:

1 [Link]()

Out[85]:

Year_Birth Income Kidhome Teenhome Recency MntWines MntFruit

count 2034.000000 2034.000000 2034.000000 2034.000000 2034.000000 2034.000000 2034.00000

mean 1968.802852 52357.791544 0.442970 0.508358 48.844641 305.180924 26.29252

std 11.975420 25526.956988 0.535914 0.546130 28.983678 337.323274 39.77288

min 1893.000000 1730.000000 0.000000 0.000000 0.000000 0.000000 0.00000

25% 1959.000000 35528.250000 0.000000 0.000000 24.000000 23.000000 2.00000

50% 1970.000000 51533.000000 0.000000 0.000000 49.000000 175.500000 8.00000

75% 1977.000000 68480.750000 1.000000 1.000000 74.000000 505.000000 33.00000

max 1996.000000 666666.000000 2.000000 2.000000 99.000000 1493.000000 199.00000

8 rows × 25 columns

 

In [86]:

1 [Link]()

Out[86]:

Year_Birth Education Marital_Status Income Kidhome Teenhome Recency MntWines MntFru

0 1957 Graduation Single 58138.0 0 0 58 635

1 1954 Graduation Single 46344.0 1 1 38 11

2 1965 Graduation Together 71613.0 0 0 26 426

3 1984 Graduation Together 26646.0 1 0 26 11

4 1981 PhD Married 58293.0 1 0 94 173

5 rows × 27 columns

 

In [87]:

1 #Sepearting categorical & numeric Data

2 df_Numeric=df.select_dtypes(include="int")
3 df_categoric=df.select_dtypes(include="object")
In [88]:

1 df_categoric.head()

Out[88]:

Education Marital_Status

0 Graduation Single

1 Graduation Single

2 Graduation Together

3 Graduation Together

4 PhD Married

In [89]:

1 #Encoding

In [90]:

1 enc_categoric=pd.get_dummies(df_categoric)
2 enc_categoric.head()

Out[90]:

Education_2n
Education_Basic Education_Graduation Education_Master Education_PhD Marital_
Cycle

0 0 0 1 0 0

1 0 0 1 0 0

2 0 0 1 0 0

3 0 0 1 0 0

4 0 0 0 0 1

 

In [91]:

1 #Concating Numeric & Encoded Categoric Data

In [92]:

1 final_df=[Link]([enc_categoric,df_Numeric],axis=1)
In [93]:

1 final_df.head()

Out[93]:

Education_2n
Education_Basic Education_Graduation Education_Master Education_PhD Marital_
Cycle

0 0 0 1 0 0

1 0 0 1 0 0

2 0 0 1 0 0

3 0 0 1 0 0

4 0 0 0 0 1

5 rows × 37 columns

 

In [94]:

1 final_df.columns

Out[94]:

Index(['Education_2n Cycle', 'Education_Basic', 'Education_Graduation',

'Education_Master', 'Education_PhD', 'Marital_Status_Absurd',
'Marital_Status_Alone', 'Marital_Status_Divorced',
'Marital_Status_Married', 'Marital_Status_Single',
'Marital_Status_Together', 'Marital_Status_Widow',
'Marital_Status_YOLO', 'Year_Birth', 'Kidhome', 'Teenhome', 'Recency',
'MntWines', 'MntFruits', 'MntMeatProducts', 'MntFishProducts',
'MntSweetProducts', 'MntGoldProds', 'NumDealsPurchases',
'NumWebPurchases', 'NumCatalogPurchases', 'NumStorePurchases',
'NumWebVisitsMonth', 'AcceptedCmp3', 'AcceptedCmp4', 'AcceptedCmp5',
'AcceptedCmp1', 'AcceptedCmp2', 'Complain', 'Z_CostContact',
'Z_Revenue', 'Response'],
dtype='object')

PCA
In [137]:

1 from [Link] import StandardScaler

2 sc=StandardScaler()
3 scaled_df=sc.fit_transform(final_df)
4 df2=[Link](scaled_df,columns=final_df.columns)

In [138]:

1 scaled_df.shape

Out[138]:

(2034, 37)
In [139]:

1 [Link]()

Out[139]:

Education_2n
Education_Basic Education_Graduation Education_Master Education_PhD Marital_
Cycle

0 -0.316313 -0.157115 0.998035 -0.444837 -0.529198

1 -0.316313 -0.157115 0.998035 -0.444837 -0.529198

2 -0.316313 -0.157115 0.998035 -0.444837 -0.529198

3 -0.316313 -0.157115 0.998035 -0.444837 -0.529198

4 -0.316313 -0.157115 -1.001969 -0.444837 1.889652

5 rows × 37 columns

 

In [140]:

1 from [Link] import PCA

In [141]:

1 pca=PCA(n_components=37)
2 pca.fit_transform(df2)

Out[141]:

array([[ 3.92126114e+00, -9.61448340e-01, -4.18362311e-01, ...,

6.81726420e-15, 4.70969941e-16, 1.84690340e-17],
[-2.36140150e+00, -5.12049484e-01, -3.04982935e-01, ...,
3.94533695e-15, 1.38987152e-17, -3.82208087e-17],
[ 1.50092229e+00, -4.27692395e-01, -1.38093928e+00, ...,
-1.53724166e-14, 7.59451335e-17, 3.11820417e-17],
...,
[ 1.50103128e+00, 2.68854853e-01, 4.08566507e-01, ...,
-3.13574979e-16, 1.56782122e-17, -9.09263745e-18],
[ 1.56188422e+00, 1.38250713e+00, -1.05948585e+00, ...,
7.86248061e-16, 1.99674898e-17, -6.51998320e-18],
[-1.78295060e+00, 1.65644879e+00, 1.32880184e+00, ...,
1.41045263e-15, -4.02049749e-18, -8.11276839e-18]])
In [142]:

1 #expalined variance ratio

2 pca.explained_variance_ratio_

Out[142]:

array([1.72192328e-01, 6.28764599e-02, 5.61055137e-02, 4.64831620e-02,

4.32938627e-02, 4.07701572e-02, 3.76059249e-02, 3.50710137e-02,
3.40452270e-02, 3.32582107e-02, 3.07888139e-02, 2.99418829e-02,
2.98386064e-02, 2.90265643e-02, 2.84050295e-02, 2.75235307e-02,
2.67648958e-02, 2.64538757e-02, 2.27422479e-02, 2.09773868e-02,
1.90604822e-02, 1.77636535e-02, 1.69051260e-02, 1.62791683e-02,
1.52097793e-02, 1.44799030e-02, 1.22756402e-02, 1.18222605e-02,
1.06760730e-02, 1.00919538e-02, 8.94729489e-03, 6.60407766e-03,
5.71989396e-03, 8.92816949e-30, 2.40570590e-30, 7.05358138e-34,
7.05358138e-34])

In [154]:

1 [Link](figsize=(10,5))
2 [Link]([Link](pca.explained_variance_ratio_),linewidth=4,markersize=10,color="blue")
3 [Link]('Number of Components')
4 [Link]('Explained Variance Ratio')
5 [Link]()
In [163]:

1 [Link](figsize=(10,5))
2 [Link]([Link](pca.explained_variance_ratio_),color='blue', linewidth=4,)
3 [Link](y=0.9, color='cyan', linestyle='--', label='90% EVR')
4 [Link](x=23, color='cyan', linestyle='--', label='Selected Components')
5 [Link](23,0.9,marker="*",linewidths=10,color="red")
6 [Link]('PCA - Explained Variance Ratio')
7 [Link]('Number of Components')
8 [Link]('Explained Variance Ratio')
9 [Link]([Link](0, 1.1, step=0.1))
10 [Link](3, 0.85, 'Selected Components(23)', color='r')
11 [Link](loc="lower right")
12 [Link]()

In [164]:

1 #23 components are showing 90% of the variance in our data,so we'll build our PCA again wi

In [165]:

1 pca=PCA(n_components=23)

In [166]:

1 df3=[Link](pca.fit_transform(df2))
In [167]:

1 [Link]()

Out[167]:

0 1 2 3 4 5 6 7 8

0 3.921261 -0.961448 -0.418362 1.552713 -0.998744 2.012662 -1.186018 0.352056 -0.528134

1 -2.361402 -0.512049 -0.304983 0.592715 -1.036934 -0.347183 -1.857775 0.830816 -0.269280

2 1.500922 -0.427692 -1.380939 0.160132 -1.317587 -0.722257 0.748517 -0.629505 -0.117337

3 -2.259714 -1.557122 -0.086753 0.581038 -1.127287 -0.954844 1.062398 -0.710251 -0.124947

4 -0.590543 0.331769 -0.096791 -0.775769 1.749521 0.541293 -0.048923 -1.439071 -1.174442

5 rows × 23 columns

 

Clustering

[Link] Clustering
In [168]:

1 from [Link] import KMeans

In [169]:

1 wcss=[]

In [170]:

1 for i in range(1,5):
2 kmean=KMeans(n_clusters=i,init="random",random_state=42)
3 [Link](df3)
4 [Link](kmean.inertia_)

In [171]:

1 print(wcss)

[63209.17068842075, 53715.05325913177, 50513.20588634547, 48277.99630918754]

In [172]:

1 [Link](figsize=(10,5))
2 [Link](range(1,5),wcss,color="blue",linewidth=3)
3 [Link]('WCSS vs Number of Clusters')
4 [Link]('Number of Clusters')
5 [Link]('WCSS')
6 [Link]()

In [209]:

1 [Link](figsize=(10,5))
2 [Link](range(1,5),wcss,color="blue",linewidth=3)
3 [Link]('WCSS vs Number of Clusters')
4 [Link](2,wcss[1],s=200,c="cyan",marker="*",linewidth=5)
5 [Link]('Number of Clusters')
6 [Link]('WCSS')
7 [Link]()
In [174]:

1 from [Link] import silhouette_score

In [175]:

1 silhouette_score_lst=[]
2
3 for i in range(2,6):
4 silhouette_score_lst.append(silhouette_score(df3,(KMeans(n_clusters=i).fit_predict(df3

In [176]:

1 silhouette_score_lst

Out[176]:

[0.18256678718476013,
0.12731079376982762,
0.12685890301831806,
0.1261638680247387]

In [212]:

1 #plotting
2 [Link](figsize=(10,5))
3
4 k=[2,3,4,5]
5 [Link](k,silhouette_score_lst,color="blue",edgecolor="red")
6
7 [Link]("Number of clusters")
8 [Link]("silhouette_score")
9 [Link]()

Elbow Graph shows that we should select 2 clusters and also silhouette scores is maximum when clusters
are 2
In [178]:

1 #Making model again with 2 clusters

In [179]:

1 kmeans=KMeans(n_clusters=2,random_state=42)

In [180]:

1 [Link](df3)

Out[180]:

KMeans(n_clusters=2, random_state=42)

In [181]:

1 pred=[Link](df3)
2 pred

Out[181]:

array([1, 0, 1, ..., 1, 1, 0])

In [213]:

1 x = [Link][:, 0]
2 y = [Link][:, 1]
3 [Link](figsize=(10,5))
4 [Link](x, y, c=pred,edgecolors="cyan")
5 [Link]('PC1')
6 [Link]('PC2')
7 [Link]('Cluster Visualization (PCA)')
8 [Link]()
In [220]:

1 x = [Link][:, 0]
2 y = [Link][:, 22]
3 [Link](figsize=(10,5))
4 [Link](x, y, c=pred,edgecolors="cyan")
5 [Link]()
In [184]:

1 # visualzing all
2 [Link](figsize=(10,5))
3
4 [Link].scatter_matrix([Link](df3), c=pred, figsize=(15,15), diagonal='hist')
5 [Link]()

<Figure size 1000x500 with 0 Axes>

In [221]:

1 #Viewing subset of it
2 dim_subset = [0, 1, 2, 4, 5,12,17,21]
3 [Link].scatter_matrix([Link]([Link][:, dim_subset]), c=pred, figsize=(15,15),
4 [Link]()
In [186]:

1 from mpl_toolkits.mplot3d import Axes3D

2
3 # Create a 3D plot with different colors for each cluster
4 fig = [Link](figsize=(8,8))
5 ax = fig.add_subplot(111, projection='3d')
6 [Link]([Link][:,0], [Link][:,1], [Link][:,2], c=pred, s=40,edgecolor="cyan")
7 ax.set_xlabel('PCA 1')
8 ax.set_ylabel('PCA 2')
9 ax.set_zlabel('PCA 3')
10 ax.set_title('K-Means Clustering Results (3D)')
11 [Link]()

[Link] Clustering
In [187]:

1 from [Link] import AgglomerativeClustering

2 import [Link] as shc

In [223]:

1 [Link](figsize=(15,12))
2 den=[Link]([Link](df3,method="ward"))
3 [Link](y=100,color="r",linestyle="--")
4 [Link]()

Hierachial is showing with 100 clusters so its definately not a good approach.

[Link]
In [189]:

1 from [Link] import DBSCAN

In [190]:

1 dbscan=DBSCAN(eps=27,min_samples=70)
2 [Link](df3)

Out[190]:

DBSCAN(eps=27, min_samples=70)
In [191]:

1 #Checking labels
2 dbscan.labels_

Out[191]:

array([0, 0, 0, ..., 0, 0, 0], dtype=int64)

In [192]:

1 #Total no of clusters
2 len(set(dbscan.labels_))

Out[192]:

In [193]:

1 from [Link] import silhouette_score

2 silhouette_score(df3,dbscan.labels_)

Out[193]:

0.7784382051017664
In [258]:

1 fig = [Link](figsize=(10, 8))

2 ax = fig.add_subplot(111, projection='3d')
3 [Link]([Link][:, 0], [Link][:, 1],[Link][:, 2], c=dbscan.labels_,edgecolor="cyan
4 ax.set_xlabel('Feature 1')
5 ax.set_ylabel('Feature 2')
6 ax.set_zlabel('Feature 3')
7 ax.set_title('DBSCAN Clustering Results (3D)')
8 [Link]()

Conclusions:
[Link] KMeans and DBSCAN were able to efficiently separate out the clusters in the dataset.

[Link], when comparing the two algorithms using the silhouette score, DBSCAN outperformed KMeans with a
score of 0.7784, while KMeans had a score of 0.182. This suggests that, despite the fact that KMeans appeared to
perform better visually, the clusters produced by DBSCAN were more distinct and better separated.

[Link], based on the silhouette score, DBSCAN is a better clustering algorithm for this dataset.
Project By:

Rohit Vyavahare

Linkedin [Link]
([Link]

Github : [Link]
([Link]