Customer Segmentation is one the most important applications of unsupervised learning. Using clustering techniques, one can identify the several segments of customers allowing them to target the potential user base.
This notebook has the following sections
- Importing Necessary Libraries
- Loading the data
- Exploratory Data Analysis
- Applying clustering Techniques
- Finding the optimal Clusters
- Applying PCA
- Find the optimal clusters
- Visualization of clusters
import numpy as np
import pandas as pd
# for standardizing the dataset
from sklearn.preprocessing import StandardScaler
# for visualization
import matplotlib.pyplot as plt
import seaborn as sns
sns.set()
# for the clustering techniques
from scipy.cluster.hierarchy import dendrogram,linkage
from sklearn.cluster import KMeansdf_users=pd.read_csv('user_side_data_raw.tsv',sep='\t',names=['id','gender','age','country','date'])df_users.head()
<style scoped>
.dataframe tbody tr th:only-of-type {
vertical-align: middle;
}
</style>
.dataframe tbody tr th {
vertical-align: top;
}
.dataframe thead th {
text-align: right;
}
| id | gender | age | country | date | |
|---|---|---|---|---|---|
| 0 | 00000c289a1829a808ac09c00daf10bc3c4e223b | f | 22.0 | Germany | Feb 1, 2007 |
| 1 | 00001411dc427966b17297bf4d69e7e193135d89 | f | NaN | Canada | Dec 4, 2007 |
| 2 | 00004d2ac9316e22dc007ab2243d6fcb239e707d | NaN | NaN | Germany | Sep 1, 2006 |
| 3 | 000063d3fe1cf2ba248b9e3c3f0334845a27a6bf | m | 19.0 | Mexico | Apr 28, 2008 |
| 4 | 00007a47085b9aab8af55f52ec8846ac479ac4fe | m | 28.0 | United States | Jan 27, 2006 |
df_users.shape(359347, 5)
df_users.isnull().sum()id 0
gender 32775
age 74900
country 0
date 0
dtype: int64
# proportions of null values
(df_users.gender.isnull().sum()/df_users.shape[0])*1009.12071062232327
(df_users.age.isnull().sum()/df_users.shape[0])*10020.84336310029025
df_users.gender.fillna(df_users.gender.mode()[0],inplace=True)df_users.age.fillna(int(df_users.age.mean()),inplace=True)df_users.isnull().sum()id 0
gender 0
age 0
country 0
date 0
dtype: int64
df_users.gender.value_counts()m 274417
f 84930
Name: gender, dtype: int64
df_users.age.mean()25.07761022076155
df_users.country.value_counts()United States 67044
Germany 31651
United Kingdom 29902
Poland 20987
Russian Federation 19833
Brazil 14534
Sweden 13122
Spain 13051
Finland 11579
Netherlands 9650
Canada 8679
France 7529
Italy 7525
Australia 7135
Japan 6637
Turkey 6452
Norway 5155
Mexico 4834
Czech Republic 4774
Ukraine 4396
Belgium 3803
Portugal 3196
Switzerland 3053
Bulgaria 2800
Austria 2796
Chile 2794
Argentina 2640
Romania 2636
Denmark 2508
Hungary 1985
...
Equatorial Guinea 10
Saint Kitts and Nevis 10
San Marino 9
Cameroon 9
Namibia 9
Senegal 9
Saint Lucia 8
Rwanda 8
Gabon 8
Comoros 8
Mayotte 8
Tonga 8
Grenada 7
Myanmar 7
Mali 7
Malawi 6
Belize 6
Eritrea 6
Guinea-Bissau 5
Central African Republic 5
Sudan 5
Guyana 5
Suriname 5
Mauritania 5
Marshall Islands 4
French Guiana 4
Palau 3
Benin 3
Liberia 3
Gambia 3
Name: country, Length: 239, dtype: int64
df_users['joined year']=df_users.date.apply(lambda x:x[-4:])df_users.head()
<style scoped>
.dataframe tbody tr th:only-of-type {
vertical-align: middle;
}
</style>
.dataframe tbody tr th {
vertical-align: top;
}
.dataframe thead th {
text-align: right;
}
| id | gender | age | country | date | joined year | |
|---|---|---|---|---|---|---|
| 0 | 00000c289a1829a808ac09c00daf10bc3c4e223b | f | 22.0 | Germany | Feb 1, 2007 | 2007 |
| 1 | 00001411dc427966b17297bf4d69e7e193135d89 | f | 25.0 | Canada | Dec 4, 2007 | 2007 |
| 2 | 00004d2ac9316e22dc007ab2243d6fcb239e707d | m | 25.0 | Germany | Sep 1, 2006 | 2006 |
| 3 | 000063d3fe1cf2ba248b9e3c3f0334845a27a6bf | m | 19.0 | Mexico | Apr 28, 2008 | 2008 |
| 4 | 00007a47085b9aab8af55f52ec8846ac479ac4fe | m | 28.0 | United States | Jan 27, 2006 | 2006 |
df_users.drop(['date'],axis=1,inplace=True)df_users.set_index(['id'],inplace=True)df_users.head()
<style scoped>
.dataframe tbody tr th:only-of-type {
vertical-align: middle;
}
</style>
.dataframe tbody tr th {
vertical-align: top;
}
.dataframe thead th {
text-align: right;
}
| gender | age | country | joined year | |
|---|---|---|---|---|
| id | ||||
| 00000c289a1829a808ac09c00daf10bc3c4e223b | f | 22.0 | Germany | 2007 |
| 00001411dc427966b17297bf4d69e7e193135d89 | f | 25.0 | Canada | 2007 |
| 00004d2ac9316e22dc007ab2243d6fcb239e707d | m | 25.0 | Germany | 2006 |
| 000063d3fe1cf2ba248b9e3c3f0334845a27a6bf | m | 19.0 | Mexico | 2008 |
| 00007a47085b9aab8af55f52ec8846ac479ac4fe | m | 28.0 | United States | 2006 |
df_seg=pd.get_dummies(df_users,['gender','country','joined year'])df_seg.head()
<style scoped>
.dataframe tbody tr th:only-of-type {
vertical-align: middle;
}
</style>
.dataframe tbody tr th {
vertical-align: top;
}
.dataframe thead th {
text-align: right;
}
| age | gender_f | gender_m | country_Afghanistan | country_Albania | country_Algeria | country_American Samoa | country_Andorra | country_Angola | country_Anguilla | ... | country_Zambia | country_Zimbabwe | joined year_2002 | joined year_2003 | joined year_2004 | joined year_2005 | joined year_2006 | joined year_2007 | joined year_2008 | joined year_2009 | |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| id | |||||||||||||||||||||
| 00000c289a1829a808ac09c00daf10bc3c4e223b | 22.0 | 1 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | ... | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 1 | 0 | 0 |
| 00001411dc427966b17297bf4d69e7e193135d89 | 25.0 | 1 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | ... | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 1 | 0 | 0 |
| 00004d2ac9316e22dc007ab2243d6fcb239e707d | 25.0 | 0 | 1 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | ... | 0 | 0 | 0 | 0 | 0 | 0 | 1 | 0 | 0 | 0 |
| 000063d3fe1cf2ba248b9e3c3f0334845a27a6bf | 19.0 | 0 | 1 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | ... | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 1 | 0 |
| 00007a47085b9aab8af55f52ec8846ac479ac4fe | 28.0 | 0 | 1 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | ... | 0 | 0 | 0 | 0 | 0 | 0 | 1 | 0 | 0 | 0 |
5 rows × 250 columns
df_users.describe(include='all')
<style scoped>
.dataframe tbody tr th:only-of-type {
vertical-align: middle;
}
</style>
.dataframe tbody tr th {
vertical-align: top;
}
.dataframe thead th {
text-align: right;
}
| gender | age | country | joined year | |
|---|---|---|---|---|
| count | 359347 | 359347.000000 | 359347 | 359347 |
| unique | 2 | NaN | 239 | 8 |
| top | m | NaN | United States | 2008 |
| freq | 274417 | NaN | 67044 | 120808 |
| mean | NaN | 25.077610 | NaN | NaN |
| std | NaN | 19.276048 | NaN | NaN |
| min | NaN | -1337.000000 | NaN | NaN |
| 25% | NaN | 21.000000 | NaN | NaN |
| 50% | NaN | 25.000000 | NaN | NaN |
| 75% | NaN | 26.000000 | NaN | NaN |
| max | NaN | 1002.000000 | NaN | NaN |
from sklearn.preprocessing import LabelEncoder
lb_encoder=LabelEncoder()countries=lb_encoder.fit_transform(df_users.country)year_encoder=LabelEncoder()
years=lb_encoder.fit_transform(df_users['joined year'])yearsarray([5, 5, 4, ..., 5, 3, 6])
gender_encoder=LabelEncoder()
genders=lb_encoder.fit_transform(df_users.gender)gendersarray([0, 0, 1, ..., 1, 1, 1])
Clustering is the task of grouping together a set of objects in a way that objects in the same cluster are more similar to each other than to objects in other clusters. Similarity is a metric that reflects the strength of relationship between two data objects. Clustering is mainly used for exploratory data mining.
we have two ways to find the optimal no.of clusters
- By Hierarchical clustering we can find the optimal no.of clusters by observing the dendogram
- By ploting the WCSS of a range of cluster using Bottom up approach
The first option crashed the kernel because of lack of enough hardware requirement in the laptop.I'm opting for second one
K-Means attempts to classify data without having first been trained with labeled data. Once the algorithm has been run and the groups are defined, any new data can be easily assigned to the most relevant group.
wcss=[]
for i in range(1,11):
kmeans=KMeans(n_clusters=i,init='k-means++',random_state=42)
kmeans.fit(df_seg)
wcss.append(kmeans.inertia_)
print(kmeans.inertia_)134258498.48275355
28492159.521132484
15087558.004872149
8737924.294237286
5680436.533094462
4150917.3284346457
3079109.5899161794
2474939.295240826
2105803.1053973585
1820427.9377313748
wcss[134258498.48275355,
28492159.521132484,
15087558.004872149,
8737924.294237286,
5680436.533094462,
4150917.3284346457,
3079109.5899161794,
2474939.295240826,
2105803.1053973585,
1820427.9377313748]
f=open('wcss.txt','w')
for i in wcss:
f.write(str(i))
f.write('\n')
f.close()plt.figure(figsize=(12,9))
plt.plot(range(1,11),wcss,marker='o',linestyle='--')
plt.xlabel('Number of clusters')
plt.ylabel('wcss')
plt.show()
We graph the relationship between the number of clusters and Within Cluster Sum of Squares (WCSS) then we select the number of clusters where the change in WCSS begins to level off (elbow method).
WCSS is defined as the sum of the squared distance between each member of the cluster and its centroid.
kmeans=KMeans(n_clusters=3,init='k-means++',random_state=42)
kmeans.fit(df_seg)KMeans(algorithm='auto', copy_x=True, init='k-means++', max_iter=300,
n_clusters=3, n_init=10, n_jobs=None, precompute_distances='auto',
random_state=42, tol=0.0001, verbose=0)
df_segm_kmeans=df_seg.copy()
df_segm_kmeans['segment K-means']=kmeans.labels_df_users.head()
<style scoped>
.dataframe tbody tr th:only-of-type {
vertical-align: middle;
}
</style>
.dataframe tbody tr th {
vertical-align: top;
}
.dataframe thead th {
text-align: right;
}
| gender | age | country | joined year | |
|---|---|---|---|---|
| id | ||||
| 00000c289a1829a808ac09c00daf10bc3c4e223b | f | 22.0 | Germany | 2007 |
| 00001411dc427966b17297bf4d69e7e193135d89 | f | 25.0 | Canada | 2007 |
| 00004d2ac9316e22dc007ab2243d6fcb239e707d | m | 25.0 | Germany | 2006 |
| 000063d3fe1cf2ba248b9e3c3f0334845a27a6bf | m | 19.0 | Mexico | 2008 |
| 00007a47085b9aab8af55f52ec8846ac479ac4fe | m | 28.0 | United States | 2006 |
df_users_en=df_users.copy()df_users_en['gender']=genders
df_users_en['country']=countries
df_users_en['joined year']=yearsdf_users_en.head()
<style scoped>
.dataframe tbody tr th:only-of-type {
vertical-align: middle;
}
</style>
.dataframe tbody tr th {
vertical-align: top;
}
.dataframe thead th {
text-align: right;
}
| gender | age | country | joined year | |
|---|---|---|---|---|
| id | ||||
| 00000c289a1829a808ac09c00daf10bc3c4e223b | 0 | 22.0 | 79 | 5 |
| 00001411dc427966b17297bf4d69e7e193135d89 | 0 | 25.0 | 37 | 5 |
| 00004d2ac9316e22dc007ab2243d6fcb239e707d | 1 | 25.0 | 79 | 4 |
| 000063d3fe1cf2ba248b9e3c3f0334845a27a6bf | 1 | 19.0 | 136 | 6 |
| 00007a47085b9aab8af55f52ec8846ac479ac4fe | 1 | 28.0 | 225 | 4 |
test_wcss=[]
for i in range(1,11):
kmeans=KMeans(n_clusters=i,init='k-means++',random_state=42)
kmeans.fit(df_users_en)
test_wcss.append(kmeans.inertia_)
print(kmeans.inertia_)2086135898.8955686
431290273.3553824
290135430.90301967
184763290.25917345
86585872.56605095
66684087.42352505
52578488.68448655
44171563.90171374
38522141.4220099
33634118.314277805
plt.figure(figsize=(12,9))
plt.plot(range(1,11),test_wcss,marker='o',linestyle='--')
plt.xlabel('Number of clusters')
plt.ylabel('wcss')
plt.show()
From the above graph we can conclude that there are three optimal clusters can be formed using Elbow-point method
kmeans=KMeans(n_clusters=3,init='k-means++',random_state=42)
kmeans.fit(df_users_en)KMeans(algorithm='auto', copy_x=True, init='k-means++', max_iter=300,
n_clusters=3, n_init=10, n_jobs=None, precompute_distances='auto',
random_state=42, tol=0.0001, verbose=0)
df_en_segm_kmeans=df_users_en.copy()
df_en_segm_kmeans['segment K-means']=kmeans.labels_df_en_segm_kmeans.head()
<style scoped>
.dataframe tbody tr th:only-of-type {
vertical-align: middle;
}
</style>
.dataframe tbody tr th {
vertical-align: top;
}
.dataframe thead th {
text-align: right;
}
| gender | age | country | joined year | segment K-means | |
|---|---|---|---|---|---|
| id | |||||
| 00000c289a1829a808ac09c00daf10bc3c4e223b | 0 | 22.0 | 79 | 5 | 0 |
| 00001411dc427966b17297bf4d69e7e193135d89 | 0 | 25.0 | 37 | 5 | 0 |
| 00004d2ac9316e22dc007ab2243d6fcb239e707d | 1 | 25.0 | 79 | 4 | 0 |
| 000063d3fe1cf2ba248b9e3c3f0334845a27a6bf | 1 | 19.0 | 136 | 6 | 2 |
| 00007a47085b9aab8af55f52ec8846ac479ac4fe | 1 | 28.0 | 225 | 4 | 1 |
df_en_segm_kmeans.groupby(['segment K-means']).count()
<style scoped>
.dataframe tbody tr th:only-of-type {
vertical-align: middle;
}
</style>
.dataframe tbody tr th {
vertical-align: top;
}
.dataframe thead th {
text-align: right;
}
| gender | age | country | joined year | |
|---|---|---|---|---|
| segment K-means | ||||
| 0 | 139132 | 139132 | 139132 | 139132 |
| 1 | 143475 | 143475 | 143475 | 143475 |
| 2 | 76740 | 76740 | 76740 | 76740 |
x_axis=df_en_segm_kmeans['age']
y_axis=df_en_segm_kmeans['gender']
plt.figure(figsize=(10,8))
sns.scatterplot(x_axis,y_axis,hue=df_en_segm_kmeans['segment K-means'])
plt.title('Segmentation K-means')
plt.show()
It is a common practice to apply PCA (principal component analysis) before a clustering algorithm (such as k-means). It is believed that it improves the clustering results in practice (noise reduction).
df_seg.head()
<style scoped>
.dataframe tbody tr th:only-of-type {
vertical-align: middle;
}
</style>
.dataframe tbody tr th {
vertical-align: top;
}
.dataframe thead th {
text-align: right;
}
| age | gender_f | gender_m | country_Afghanistan | country_Albania | country_Algeria | country_American Samoa | country_Andorra | country_Angola | country_Anguilla | ... | country_Zambia | country_Zimbabwe | joined year_2002 | joined year_2003 | joined year_2004 | joined year_2005 | joined year_2006 | joined year_2007 | joined year_2008 | joined year_2009 | |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| id | |||||||||||||||||||||
| 00000c289a1829a808ac09c00daf10bc3c4e223b | 22.0 | 1 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | ... | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 1 | 0 | 0 |
| 00001411dc427966b17297bf4d69e7e193135d89 | 25.0 | 1 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | ... | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 1 | 0 | 0 |
| 00004d2ac9316e22dc007ab2243d6fcb239e707d | 25.0 | 0 | 1 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | ... | 0 | 0 | 0 | 0 | 0 | 0 | 1 | 0 | 0 | 0 |
| 000063d3fe1cf2ba248b9e3c3f0334845a27a6bf | 19.0 | 0 | 1 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | ... | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 1 | 0 |
| 00007a47085b9aab8af55f52ec8846ac479ac4fe | 28.0 | 0 | 1 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | ... | 0 | 0 | 0 | 0 | 0 | 0 | 1 | 0 | 0 | 0 |
5 rows × 250 columns
df_seg.shape(359347, 250)
from sklearn.decomposition import PCA
pca=PCA()
pca.fit(df_seg)PCA(copy=True, iterated_power='auto', n_components=None, random_state=None,
svd_solver='auto', tol=0.0, whiten=False)
On the applying the pca on the dataset we can get the variance explained by each column/field. As a rule of thumb we have to retain the components explaining more than 80% variance
pca.explained_variance_ratio_array([9.94508186e-01, 9.73522817e-04, 8.20579773e-04, 5.95721570e-04,
4.23043046e-04, 3.16620110e-04, 2.36815861e-04, 2.24564767e-04,
1.94659919e-04, 1.52096641e-04, 1.32009525e-04, 1.04611178e-04,
9.71849613e-05, 9.02506282e-05, 7.90944697e-05, 6.88602904e-05,
6.24582865e-05, 5.91033460e-05, 5.59826220e-05, 5.40947256e-05,
5.09339156e-05, 4.85152270e-05, 4.18295955e-05, 3.72284667e-05,
3.57277045e-05, 3.34830053e-05, 2.97241284e-05, 2.62015022e-05,
2.33735151e-05, 2.21352569e-05, 2.08321749e-05, 2.08007512e-05,
2.00785538e-05, 1.96385456e-05, 1.88673504e-05, 1.61374506e-05,
1.47573341e-05, 1.44643374e-05, 1.40240498e-05, 1.37709835e-05,
1.31064515e-05, 1.23692322e-05, 1.18111175e-05, 1.10000004e-05,
1.00279260e-05, 9.47876445e-06, 9.14714313e-06, 8.86760541e-06,
8.67800866e-06, 7.70307208e-06, 7.67159210e-06, 5.76894105e-06,
5.19930343e-06, 4.80708919e-06, 4.56295442e-06, 4.46261434e-06,
4.26692383e-06, 3.83936073e-06, 3.67754963e-06, 3.34438438e-06,
3.18773258e-06, 3.12815711e-06, 2.87014131e-06, 2.75656584e-06,
2.72712673e-06, 2.47237075e-06, 2.38166637e-06, 2.19906353e-06,
2.05900936e-06, 1.97978608e-06, 1.90307330e-06, 1.68047567e-06,
1.59470740e-06, 1.50658856e-06, 1.46615802e-06, 1.33682246e-06,
1.27957460e-06, 1.07718066e-06, 9.98022358e-07, 9.68251415e-07,
9.57905326e-07, 9.19400987e-07, 8.45930243e-07, 7.94271484e-07,
7.54658274e-07, 7.44790552e-07, 7.20438788e-07, 6.95982213e-07,
6.74934359e-07, 6.59264476e-07, 6.50888084e-07, 6.25582753e-07,
6.10743920e-07, 5.90678072e-07, 5.73134299e-07, 5.59405831e-07,
5.51150575e-07, 5.40057633e-07, 5.20446256e-07, 5.04597269e-07,
4.87954010e-07, 4.76684985e-07, 4.71534318e-07, 4.63970861e-07,
4.38285746e-07, 4.16822315e-07, 3.99990982e-07, 3.95251715e-07,
3.89967669e-07, 3.82741298e-07, 3.75220497e-07, 3.67580918e-07,
3.56091579e-07, 3.47480286e-07, 3.42618219e-07, 3.42610019e-07,
3.30604562e-07, 3.26406772e-07, 3.20276480e-07, 3.06736254e-07,
2.97923602e-07, 2.93015723e-07, 2.85970480e-07, 2.76959634e-07,
2.68104924e-07, 2.64372514e-07, 2.58769842e-07, 2.53241597e-07,
2.53223936e-07, 2.49628719e-07, 2.45792638e-07, 2.45707335e-07,
2.40951568e-07, 2.36566801e-07, 2.30896537e-07, 2.30892295e-07,
2.30889871e-07, 2.30885957e-07, 2.26742891e-07, 2.23448883e-07,
2.23447035e-07, 2.23439483e-07, 2.19663069e-07, 2.16001219e-07,
2.16000859e-07, 2.15998211e-07, 2.15996978e-07, 2.15995929e-07,
2.15979867e-07, 2.09163783e-07, 2.02775823e-07, 2.01098199e-07,
1.94860441e-07, 1.89790492e-07, 1.86207356e-07, 1.86205351e-07,
1.86196977e-07, 1.79897678e-07, 1.75267874e-07, 1.71310697e-07,
1.71305824e-07, 1.71303737e-07, 1.67462478e-07, 1.63862863e-07,
1.63862532e-07, 1.63860469e-07, 1.63858676e-07, 1.63857267e-07,
1.63857164e-07, 1.58171815e-07, 1.56413481e-07, 1.56410913e-07,
1.56410211e-07, 1.50704835e-07, 1.48965558e-07, 1.48965319e-07,
1.48959351e-07, 1.42463851e-07, 1.41512222e-07, 1.35211289e-07,
1.34064397e-07, 1.28641246e-07, 1.26620115e-07, 1.26619194e-07,
1.26529159e-07, 1.20225344e-07, 1.19170003e-07, 1.13592850e-07,
1.11723423e-07, 1.11721851e-07, 1.07657805e-07, 1.04276435e-07,
1.04275964e-07, 1.04275338e-07, 1.04274754e-07, 1.04273034e-07,
1.04271979e-07, 1.04267122e-07, 1.04266543e-07, 9.86215397e-08,
9.68281310e-08, 9.68280733e-08, 9.68274111e-08, 9.68261044e-08,
9.68248163e-08, 9.68229630e-08, 9.01822837e-08, 8.93774534e-08,
8.93753397e-08, 8.34459092e-08, 8.19308868e-08, 8.19304574e-08,
8.19298303e-08, 8.19292872e-08, 8.19169571e-08, 7.53686097e-08,
7.44816457e-08, 7.44791372e-08, 7.43903163e-08, 6.79021000e-08,
6.70341310e-08, 6.70330907e-08, 6.70325489e-08, 6.06907254e-08,
5.95865186e-08, 5.95863903e-08, 5.95858414e-08, 5.95854838e-08,
5.95849659e-08, 5.26219570e-08, 5.21376438e-08, 5.21365906e-08,
4.51453399e-08, 4.46894504e-08, 4.46888105e-08, 3.79871192e-08,
3.72415398e-08, 3.72414852e-08, 3.72413734e-08, 3.72403576e-08,
3.72401695e-08, 2.99931770e-08, 2.97931632e-08, 2.26592117e-08,
2.23449135e-08, 2.23447407e-08, 2.23443392e-08, 3.07797804e-29,
4.13256137e-31, 1.35450131e-31])
plt.figure(figsize=(12,9))
plt.plot(range(1,50),pca.explained_variance_ratio_.cumsum()[:49],marker='o',linestyle='--')
plt.title('Explained variance by components')
plt.xlabel('No.of components')
plt.ylabel('Explained variance')
plt.show()
From the above graph we can see that the 10 components are explaining more than 98% variance itself
pca=PCA(n_components=10)
pca.fit(df_seg)PCA(copy=True, iterated_power='auto', n_components=10, random_state=None,
svd_solver='auto', tol=0.0, whiten=False)
scores_pca=pca.transform(df_seg)scores_pcaarray([[ 3.07924934e+00, -9.65742459e-01, 8.40539686e-01, ...,
7.06646168e-01, 3.95826494e-01, 7.82931891e-03],
[ 7.93175181e-02, -9.83262193e-01, 8.53291430e-01, ...,
-2.34652219e-03, -9.89962221e-02, -5.87799257e-02],
[ 7.68214240e-02, 4.37602123e-01, 8.39268432e-02, ...,
6.78797567e-01, 3.46492084e-01, 2.94501037e-02],
...,
[ 5.07686694e+00, 3.86188080e-01, 6.60299973e-01, ...,
-7.11075126e-02, 1.01004688e-01, 7.92880964e-03],
[ 5.07701841e+00, 3.97111786e-01, 8.15172939e-03, ...,
3.78636281e-01, -5.27255987e-01, -6.20991725e-01],
[ 4.07736262e+00, 2.01415376e-01, -7.59616724e-01, ...,
-4.35712057e-02, -8.14134376e-02, -1.57825945e-02]])
pca_wcss=[]
for i in range(1,11):
kmeans_pca=KMeans(n_clusters=i,init='k-means++',random_state=42)
kmeans_pca.fit(scores_pca)
pca_wcss.append(kmeans_pca.inertia_)
print(kmeans_pca.inertia_)134049835.27223751
28283517.23694611
14868999.474239191
8548741.72116922
5472147.923170721
3942546.1647670097
2881208.2835925072
2260121.1429546513
1886633.3945844688
1612138.7509535719
plt.figure(figsize=(12,9))
plt.plot(range(1,11),pca_wcss,marker='o',linestyle='--')
plt.xlabel('components')
plt.ylabel('wcss')
plt.show()
From the above graph we can conclude to take optimal no.of clusters to be 3 which supports the Kmeans without PCA and as well as without LabelEncoding
kmeans_pca=KMeans(n_clusters=3,init='k-means++',random_state=42)kmeans_pca.fit(scores_pca)KMeans(algorithm='auto', copy_x=True, init='k-means++', max_iter=300,
n_clusters=3, n_init=10, n_jobs=None, precompute_distances='auto',
random_state=42, tol=0.0001, verbose=0)
df_users_en.head()
<style scoped>
.dataframe tbody tr th:only-of-type {
vertical-align: middle;
}
</style>
.dataframe tbody tr th {
vertical-align: top;
}
.dataframe thead th {
text-align: right;
}
| gender | age | country | joined year | |
|---|---|---|---|---|
| id | ||||
| 00000c289a1829a808ac09c00daf10bc3c4e223b | 0 | 22.0 | 79 | 5 |
| 00001411dc427966b17297bf4d69e7e193135d89 | 0 | 25.0 | 37 | 5 |
| 00004d2ac9316e22dc007ab2243d6fcb239e707d | 1 | 25.0 | 79 | 4 |
| 000063d3fe1cf2ba248b9e3c3f0334845a27a6bf | 1 | 19.0 | 136 | 6 |
| 00007a47085b9aab8af55f52ec8846ac479ac4fe | 1 | 28.0 | 225 | 4 |
df_segm_kmeans_pca=df_users.copy()
df_segm_kmeans_pca['segment K-means']=kmeans_pca.labels_df_segm_kmeans_pca.head()
<style scoped>
.dataframe tbody tr th:only-of-type {
vertical-align: middle;
}
</style>
.dataframe tbody tr th {
vertical-align: top;
}
.dataframe thead th {
text-align: right;
}
| gender | age | country | joined year | segment K-means | |
|---|---|---|---|---|---|
| id | |||||
| 00000c289a1829a808ac09c00daf10bc3c4e223b | f | 22.0 | Germany | 2007 | 0 |
| 00001411dc427966b17297bf4d69e7e193135d89 | f | 25.0 | Canada | 2007 | 0 |
| 00004d2ac9316e22dc007ab2243d6fcb239e707d | m | 25.0 | Germany | 2006 | 0 |
| 000063d3fe1cf2ba248b9e3c3f0334845a27a6bf | m | 19.0 | Mexico | 2008 | 0 |
| 00007a47085b9aab8af55f52ec8846ac479ac4fe | m | 28.0 | United States | 2006 | 0 |
df_segm_kmeans_pca=pd.concat([df_segm_kmeans_pca.reset_index(drop=True),pd.DataFrame(scores_pca)],axis=1)df_segm_kmeans_pca.columns.values[-10:]=['component_'+str(i) for i in range(1,11)]df_segm_kmeans_pca['Labels']=df_segm_kmeans_pca['segment K-means'].map({
0:"cluster_1",
1:"cluster_2",
2:"cluster_3"
})df_segm_kmeans_pca.head()
<style scoped>
.dataframe tbody tr th:only-of-type {
vertical-align: middle;
}
</style>
.dataframe tbody tr th {
vertical-align: top;
}
.dataframe thead th {
text-align: right;
}
| gender | age | country | joined year | segment K-means | component_1 | component_2 | component_3 | component_4 | component_5 | component_6 | component_7 | component_8 | component_9 | component_10 | Labels | |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| 0 | f | 22.0 | Germany | 2007 | 0 | 3.079249 | -0.965742 | 0.840540 | -0.434122 | -0.271597 | -0.097946 | -0.395199 | 0.706646 | 0.395826 | 0.007829 | cluster_1 |
| 1 | f | 25.0 | Canada | 2007 | 0 | 0.079318 | -0.983262 | 0.853291 | -0.389060 | -0.142788 | -0.104055 | 0.059069 | -0.002347 | -0.098996 | -0.058780 | cluster_1 |
| 2 | m | 25.0 | Germany | 2006 | 0 | 0.076821 | 0.437602 | 0.083927 | 0.725797 | -0.504299 | -0.281414 | -0.455975 | 0.678798 | 0.346492 | 0.029450 | cluster_1 |
| 3 | m | 19.0 | Mexico | 2008 | 0 | 6.077364 | 0.200957 | -0.766265 | -0.276504 | -0.052819 | -0.101826 | 0.002710 | -0.052166 | -0.080768 | -0.017021 | cluster_1 |
| 4 | m | 28.0 | United States | 2006 | 0 | -2.923463 | 0.394624 | 0.137001 | 0.997569 | 0.572417 | -0.371891 | -0.118602 | -0.063246 | 0.063210 | 0.007534 | cluster_1 |
x_axis=df_segm_kmeans_pca['component_1']
y_axis=df_segm_kmeans_pca['component_10']
plt.figure(figsize=(10,8))
sns.scatterplot(x_axis,y_axis,hue=df_segm_kmeans_pca['Labels'])
plt.title('Segmentation K-means PCA')
plt.show()
import pickle
pickle.dump(pca,open('pca.pickle','wb'))
pickle.dump(kmeans_pca,open('kmeans_pca.pickle','wb'))df_segm_kmeans_pca.to_csv('user_segm_kmeans_pca.csv',index=False)