forked from adi0509/Python
-
Notifications
You must be signed in to change notification settings - Fork 0
/
pca.py
293 lines (213 loc) · 9.03 KB
/
pca.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
# -*- coding: utf-8 -*-
"""PCA.ipynb
Automatically generated by Colaboratory.
Original file is located at
https://colab.research.google.com/drive/1I9XEOKdHWUzXXtmHM1hjvbZzq162lv5i
"""
'''
resources: https://www.statisticshowto.com/sensitivity-vs-specificity-statistics/
https://en.wikipedia.org/wiki/Precision_and_recall#Precision
https://builtin.com/data-science/step-step-explanation-principal-component-analysis
https://jakevdp.github.io/PythonDataScienceHandbook/05.09-principal-component-analysis.html
https://towardsdatascience.com/pca-using-python-scikit-learn-e653f8989e60
https://gerardnico.com/data_mining/pca
https://github.com/WillKoehrsen/feature-selector/blob/master/Feature%20Selector%20Development.ipynb
https://towardsdatascience.com/a-feature-selection-tool-for-machine-learning-in-python-b64dd23710f0
https://towardsdatascience.com/churn-prediction-3a4a36c2129a
https://realpython.com/logistic-regression-python/
https://medium.com/@thevie/fighting-telco-customer-churn-problem-a-data-driven-analysis-e7c61cfae0dd
https://towardsdatascience.com/churn-prediction-770d6cb582a5
RFE: https://towardsdatascience.com/predict-employee-turnover-with-python-da4975588aa3
https://medium.com/@thevie/fighting-telco-customer-churn-problem-a-data-driven-analysis-e7c61cfae0dd
'''
# Commented out IPython magic to ensure Python compatibility.
#Import the library
import pandas as pd
import numpy as np
import sklearn
import matplotlib.pyplot as plt
import seaborn as sns
# %matplotlib inline
#Load the data set
from google.colab import files
uploaded = files.upload()
#Load the data into the data frame
df = pd.read_csv('WA_Fn-UseC_-Telco-Customer-Churn.csv')
df[df.TotalCharges == ' ']
#Note Churn = 0 = No, Churn = Yes = 1
df.head()
df[df.Churn == 'Yes'].shape
df[df.Churn == 'No'].shape
#Transform non-numeric columns into numerical columns
from sklearn.preprocessing import LabelEncoder
conv_df = df
for column in conv_df.columns:
if conv_df[column].dtype == np.number:
continue
conv_df[column] = LabelEncoder().fit_transform(conv_df[column])
cleaned_data = conv_df
#Show the first 5 rows of the new data set
cleaned_data.head()
#Scale the cleaned data
from sklearn.preprocessing import StandardScaler
X = cleaned_data.drop('Churn', axis = 1)
y = cleaned_data['Churn']
# Standardizing/scaling the features
X = StandardScaler().fit_transform(X)
'''
To determine which features or services discriminate retained cutomers from churned customers,
I will create histograms of three features (MonthlyCharges, TotalCharges, and tenure ). From the image I
can see that a high proportion of retained customers have a monthly charge between $20 and $30. ,
while the majority of churned customers had a monthly charge between $50 and $100 .
I would expect the total charges to be heavily skewed to the left since most customers have a lower monthly charge, but I
can't truly see any discrimination from the total charges plot.
The tenure histogram is heavily skewed towards the left for customers that churned and heavily skewed towards the right for customers
that were retained. We can observe that the greater the tenure, the less chance of a customer churning.
'''
import matplotlib.pyplot as plt
plt.style.use('fivethirtyeight')
numerical_features = ['tenure', 'MonthlyCharges', 'TotalCharges']
fig, ax = plt.subplots(1, 3, figsize=(28, 8))
cleaned_data[cleaned_data.Churn == 0][numerical_features].hist(bins=20, color="blue", alpha=0.5, ax=ax)
cleaned_data[cleaned_data.Churn == 1][numerical_features].hist(bins=20, color="red", alpha=0.5, ax=ax)
#PCA
from sklearn.decomposition import PCA
pca = PCA(n_components=3) #Reduce the dimensionality to 3
principalComponents = pca.fit_transform(X)
principalComponents
#Here, we can see that about 79.6% of the total variance comes from the first 2 components
ex_variance=np.var(principalComponents,axis=0)
ex_variance_ratio = ex_variance/np.sum(ex_variance)
print(ex_variance_ratio )
#PCA
from sklearn.decomposition import PCA
pca = PCA(n_components=2) #Reduce the dimensionality to 2
principalComponents = pca.fit_transform(X)
principalDf = pd.DataFrame(data = principalComponents
, columns = ['principal component 1', 'principal component 2'])
finalDf = pd.concat([principalDf, df[['Churn']]], axis = 1)
finalDf
"""
Now, since the PCA components are orthogonal to each other and they are not
correlated, we can expect to see Churn and Retained (Not Churned) classes distinctly.
Below is a plot of customer churn based on the first 2 principal components of the feature data
"""
import matplotlib.pyplot as plt
import seaborn as sns
plt.figure(figsize = (25, 12))
sns.scatterplot("principal component 1", "principal component 2", data = finalDf, hue ='Churn' )
import matplotlib.pyplot as plt
plt.style.use('fivethirtyeight')
plt.figure(figsize = (25, 12))
projected = principalComponents
num_seq = range(0, X.shape[0])
plt.scatter(projected[:, 0], projected[:, 1],
edgecolor='none', alpha=0.5, c=num_seq,
cmap=plt.cm.get_cmap('Accent', 10))
plt.xlabel('component 1')
plt.ylabel('component 2')
plt.colorbar();
#Here, we can see that 100% of the total variance comes from these 2 components
ex_variance=np.var(principalComponents,axis=0)
ex_variance_ratio = ex_variance/np.sum(ex_variance)
print(ex_variance_ratio )
principalComponents
"""
Now, since the PCA components are orthogonal to each other and they are not
correlated, we can expect to see Churn and Retained (Not Churn) classes distinctly.
Below is a plot of customer churn based on the first 2 principal components of the feature data
"""
Xax=principalComponents[:,0]
Yax=principalComponents[:,1]
labels=df['Churn']
cdict={0:'red',1:'blue'}
labl={0:'Retained',1:'Churned'}
marker={0:'*',1:'o'}
alpha={0:.3, 1:.5}
fig,ax=plt.subplots(figsize=(7,5))
fig.patch.set_facecolor('white')
for l in np.unique(labels):
ix=np.where(labels==l)
ax.scatter(Xax[ix],Yax[ix],c=cdict[l],s=40,
label=labl[l],marker=marker[l],alpha=alpha[l])
# for loop ends
plt.xlabel("First Principal Component",fontsize=14)
plt.ylabel("Second Principal Component",fontsize=14)
plt.legend()
plt.show()
# please check the scatter plot of the remaining component and you will understand the difference
from sklearn.model_selection import train_test_split
#Split the data into 80% training and 20% testing
x_train, x_test, y_train, y_test = train_test_split(X, y, test_size=0.2, random_state=42)
x_train.shape
# Make an instance of the Model / Choose the minimum number of principal components such that 95% of the variance is retained
from sklearn.decomposition import PCA
pca = PCA(.95)
#Fit PCA on training set
pca.fit(x_train)
pca.n_components_
#Apply the mapping (transform) to both the training set and the test set.
x_train = pca.transform(x_train)
x_test = pca.transform(x_test)
x_train.shape
x_train
from sklearn.linear_model import LogisticRegression
#Train the model
model = LogisticRegression()
model.fit(x_train, y_train) #Training the model
x_train.shape[1]
#recursive feature elimination
from sklearn.feature_selection import RFE
logreg = LogisticRegression()
rfe = RFE(logreg, x_train.shape[1])
rfe = rfe.fit(x_train,y_train)
print(rfe.support_)
print(rfe.ranking_)
from sklearn.metrics import classification_report
from sklearn.metrics import accuracy_score
predictions = model.predict(x_test)
print(predictions)# printing predictions
print()# Printing new line
#Check precision, recall, f1-score
print( classification_report(y_test, predictions) )
print( accuracy_score(y_test, predictions))
import numpy as np
y_test_array = np.asarray(y_test)
num_retained = 0
num_churned =0
#Note: Churn = No = 0, Churn = Yes = 1
for i in range(0,len(y_test_array)):
if y_test_array[i] == 0:
num_retained = num_retained + 1
else:
num_churned = num_churned + 1
print("The number of retained customers in the test data set: ",num_retained)
print("The number of churned customers in the test data set: ", num_churned)
print("Guessing that all customers will be retained accuracy score : ", num_retained/(num_retained+num_churned) )
#From the confusion matrix we can see that the logistic regression model predicted:
#1. True Positive = 941 (a predicted positive result that was actually positive)
#2. True Negative = 201 (a predicted negative result that was actually negative)
#3. False Positive = 172 (a predicted positive result that was actually negative)
#4. False Negative = 95 (a predicted negative result that was actually positive)
from sklearn.metrics import confusion_matrix
confusion_matrix(y_test, predictions)
# Get the confusion matrix (TP, FP, FN, & TN)
from sklearn.metrics import confusion_matrix
cm = confusion_matrix(y_test, predictions)
TP = cm[0][0]
FN = cm[0][1]
FP = cm[1][0]
TN = cm[1][1]
#Compute the miss rate
miss_rate = FN / (FN + TP)
print('The miss rate:', miss_rate)
# Get the confusion matrix (TP, FP, FN, & TN)
from sklearn.metrics import confusion_matrix
cm = confusion_matrix(y_test, predictions)
TP = cm[0][0]
FN = cm[0][1]
FP = cm[1][0]
TN = cm[1][1]
#Compute the specificity
specificity = TN / (TN + FP)
print('The specificity:', specificity)