Scribd
Upload
30 views7 pages

Data Analysis on Customer Churn

The document contains code to analyze a dataset on customer churn from a CSV file. It loads necessary libraries, reads in the data, and performs some initial exploration of the data including viewing the head, tail, shape, describe and column names. It contains over 10,000 rows of customer data including attributes like customer ID, credit score, geography, age, tenure, balance and whether they exited or not.

Uploaded by

Ajinkya Somawanshi
Copyright
© All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
30 views7 pages

Data Analysis on Customer Churn

The document contains code to analyze a dataset on customer churn from a CSV file. It loads necessary libraries, reads in the data, and performs some initial exploration of the data including viewing the head, tail, shape, describe and column names. It contains over 10,000 rows of customer data including attributes like customer ID, credit score, geography, age, tenure, balance and whether they exited or not.

Uploaded by

Ajinkya Somawanshi
Copyright
© All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd

10/29/23, 11:00 PM lab3.

ipynb - Colaboratory

IMPORT LIBRARIES

1 import pandas as pd
2 import numpy as np
3 import seaborn as sns
4 import [Link] as plt

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

1 [Link]()

account_circle RowNumber CustomerId Surname CreditScore Geography Gender Age Tenure Bala

0 1 15634602 Hargrave 619 France Female 42 2

1 2 15647311 Hill 608 Spain Female 41 1 8380

2 3 15619304 Onio 502 France Female 42 8 15966

3 4 15701354 Boni 699 France Female 39 1

4 5 15737888 Mitchell 850 Spain Female 43 2 12551

1 [Link]

(10000, 14)

1 [Link]()

RowNumber CustomerId CreditScore Age Tenure Balan

count 10000.00000 1.000000e+04 10000.000000 10000.000000 10000.000000 10000.00000

mean 5000.50000 1.569094e+07 650.528800 38.921800 5.012800 76485.88928

std 2886.89568 7.193619e+04 96.653299 10.487806 2.892174 62397.40520

min 1.00000 1.556570e+07 350.000000 18.000000 0.000000 0.00000

25% 2500.75000 1.562853e+07 584.000000 32.000000 3.000000 0.00000

50% 5000.50000 1.569074e+07 652.000000 37.000000 5.000000 97198.54000

75% 7500.25000 1.575323e+07 718.000000 44.000000 7.000000 127644.24000

max 10000.00000 1.581569e+07 850.000000 92.000000 10.000000 250898.09000

1 [Link]

Index(['RowNumber', 'CustomerId', 'Surname', 'CreditScore', 'Geography',


'Gender', 'Age', 'Tenure', 'Balance', 'NumOfProducts', 'HasCrCard',
'IsActiveMember', 'EstimatedSalary', 'Exited'],
dtype='object')

1 df

RowNumber CustomerId Surname CreditScore Geography Gender Age Tenure

0 1 15634602 Hargrave 619 France Female 42 2

1 2 15647311 Hill 608 Spain Female 41 1

2 3 15619304 Onio 502 France Female 42 8 1

3 4 15701354 Boni 699 France Female 39 1

4 5 15737888 Mitchell 850 Spain Female 43 2 1

... ... ... ... ... ... ... ... ...

9995 9996 15606229 Obijiaku 771 France Male 39 5

9996 9997 15569892 Johnstone 516 France Male 35 10

9997 9998 15584532 Liu 709 France Female 36 7

9998 9999 15682355 Sabbatini 772 Germany Male 42 3

9999 10000 15628319 Walker 792 France Female 28 4 1

10000 rows × 14 columns

[Link] 1/7
10/29/23, 11:00 PM [Link] - Colaboratory

1 [Link]()

RowNumber CustomerId Surname CreditScore Geography Gender Age Tenure Bala

0 1 15634602 Hargrave 619 France Female 42 2

1 2 15647311 Hill 608 Spain Female 41 1 8380

2 3 15619304 Onio 502 France Female 42 8 15966

3 4 15701354 Boni 699 France Female 39 1

4 5 15737888 Mitchell 850 Spain Female 43 2 12551

1 [Link]()

RowNumber CustomerId Surname CreditScore Geography Gender Age Tenure

9995 9996 15606229 Obijiaku 771 France Male 39 5

9996 9997 15569892 Johnstone 516 France Male 35 10

9997 9998 15584532 Liu 709 France Female 36 7

9998 9999 15682355 Sabbatini 772 Germany Male 42 3

9999 10000 15628319 Walker 792 France Female 28 4 1

1 [Link]

<bound method [Link] of RowNumber CustomerId Surname CreditScore Geography Gender Age \
0 1 15634602 Hargrave 619 France Female 42
1 2 15647311 Hill 608 Spain Female 41
2 3 15619304 Onio 502 France Female 42
3 4 15701354 Boni 699 France Female 39
4 5 15737888 Mitchell 850 Spain Female 43
... ... ... ... ... ... ... ...
9995 9996 15606229 Obijiaku 771 France Male 39
9996 9997 15569892 Johnstone 516 France Male 35
9997 9998 15584532 Liu 709 France Female 36
9998 9999 15682355 Sabbatini 772 Germany Male 42
9999 10000 15628319 Walker 792 France Female 28

Tenure Balance NumOfProducts HasCrCard IsActiveMember \


0 2 0.00 1 1 1
1 1 83807.86 1 0 1
2 8 159660.80 3 1 0
3 1 0.00 2 0 0
4 2 125510.82 1 1 1
... ... ... ... ... ...
9995 5 0.00 2 1 0
9996 10 57369.61 1 1 1
9997 7 0.00 1 0 1
9998 3 75075.31 2 1 0
9999 4 130142.79 1 1 0

EstimatedSalary Exited
0 101348.88 1
1 112542.58 0
2 113931.57 1
3 93826.63 0
4 79084.10 0
... ... ...
9995 96270.64 0
9996 101699.77 0
9997 42085.58 1
9998 92888.52 1
9999 38190.78 0

[10000 rows x 14 columns]>

1 [Link]

<bound method [Link] of RowNumber CustomerId Surname CreditScore Geography Gender Age \
0 1 15634602 Hargrave 619 France Female 42
1 2 15647311 Hill 608 Spain Female 41
2 3 15619304 Onio 502 France Female 42
3 4 15701354 Boni 699 France Female 39
4 5 15737888 Mitchell 850 Spain Female 43
... ... ... ... ... ... ... ...
9995 9996 15606229 Obijiaku 771 France Male 39
9996 9997 15569892 Johnstone 516 France Male 35
9997 9998 15584532 Liu 709 France Female 36
9998 9999 15682355 Sabbatini 772 Germany Male 42
9999 10000 15628319 Walker 792 France Female 28

Tenure Balance NumOfProducts HasCrCard IsActiveMember \

[Link] 2/7
10/29/23, 11:00 PM [Link] - Colaboratory
0 2 0.00 1 1 1
1 1 83807.86 1 0 1
2 8 159660.80 3 1 0
3 1 0.00 2 0 0
4 2 125510.82 1 1 1
... ... ... ... ... ...
9995 5 0.00 2 1 0
9996 10 57369.61 1 1 1
9997 7 0.00 1 0 1
9998 3 75075.31 2 1 0
9999 4 130142.79 1 1 0

EstimatedSalary Exited
0 101348.88 1
1 112542.58 0
2 113931.57 1
3 93826.63 0
4 79084.10 0
... ... ...
9995 96270.64 0
9996 101699.77 0
9997 42085.58 1
9998 92888.52 1
9999 38190.78 0

[10000 rows x 14 columns]>

1 x=[Link](['Geography', 'Gender', 'RowNumber', 'Gender', 'Exited','Surname'], axis=1)


2 x

CustomerId CreditScore Age Tenure Balance NumOfProducts HasCrCard IsAct

0 15634602 619 42 2 0.00 1 1

1 15647311 608 41 1 83807.86 1 0

2 15619304 502 42 8 159660.80 3 1

3 15701354 699 39 1 0.00 2 0

4 15737888 850 43 2 125510.82 1 1

... ... ... ... ... ... ... ...

9995 15606229 771 39 5 0.00 2 1

9996 15569892 516 35 10 57369.61 1 1

9997 15584532 709 36 7 0.00 1 0

9998 15682355 772 42 3 75075.31 2 1

9999 15628319 792 28 4 130142.79 1 1

10000 rows × 9 columns

1 [Link]()

RowNumber CustomerId Surname CreditScore Geography Gender Age Tenure Bala

0 1 15634602 Hargrave 619 France Female 42 2

1 2 15647311 Hill 608 Spain Female 41 1 8380

2 3 15619304 Onio 502 France Female 42 8 15966

3 4 15701354 Boni 699 France Female 39 1

4 5 15737888 Mitchell 850 Spain Female 43 2 12551

1 y=df["Exited"]
2

1 y

0 1
1 0
2 1
3 0
4 0
..
9995 0
9996 0
9997 1
9998 1
9999 0
Name: Exited, Length: 10000, dtype: int64

[Link] 3/7
10/29/23, 11:00 PM [Link] - Colaboratory

1 [Link](x=y)

<Axes: xlabel='Exited', ylabel='count'>

Feature Scaling

1 from [Link] import MinMaxScaler


2 o = MinMaxScaler()

1 X_scale=o.fit_transform(x)
2 X_scale

array([[0.27561613, 0.538 , 0.32432432, ..., 1. , 1. ,


0.50673489],
[0.32645436, 0.516 , 0.31081081, ..., 0. , 1. ,
0.56270874],
[0.21442143, 0.304 , 0.32432432, ..., 1. , 0. ,
0.56965435],
...,
[0.07532731, 0.718 , 0.24324324, ..., 0. , 1. ,
0.21039009],
[0.46663653, 0.844 , 0.32432432, ..., 1. , 0. ,
0.46442905],
[0.25048302, 0.884 , 0.13513514, ..., 1. , 0. ,
0.19091423]])

Cross Validation

1 from sklearn.model_selection import train_test_split


2 X_train,X_test,Y_train,Y_test = train_test_split(X_scale,y, random_state=2,test_size=0.25)

1 X_train.shape

(7500, 9)

1 X_test.shape

(2500, 9)

Initialize and build the model

1 from sklearn.neural_network import MLPClassifier


2 ann=MLPClassifier(hidden_layer_sizes=(100,100,100), max_iter=100,activation="relu", random_state=2)

1 [Link](X_train,Y_train)

[Link] 4/7
10/29/23, 11:00 PM [Link] - Colaboratory

C:\ProgramData\anaconda3\lib\site-packages\sklearn\neural_network\_multilayer_percept
[Link](
▾ MLPClassifier
MLPClassifier(hidden_layer_sizes=(100, 100, 100), max_iter=100, random_state=2)
1 y_pred=[Link](X_test)

1 y_pred

array([0, 0, 0, ..., 0, 0, 1], dtype=int64)

1 from [Link] import ConfusionMatrixDisplay,accuracy_score,classification_report

1 accuracy_score(Y_test,y_pred)

0.8424

1 print(classification_report(Y_test,y_pred))

precision recall f1-score support

0 0.87 0.94 0.91 2015


1 0.64 0.43 0.51 485

accuracy 0.84 2500


macro avg 0.76 0.68 0.71 2500
weighted avg 0.83 0.84 0.83 2500

1 ConfusionMatrixDisplay.from_predictions(Y_test,y_pred)

<[Link]._plot.confusion_matrix.ConfusionMatrixDisplay at 0x2201a10c6d0>

1 !pip install imbalanced-learn

Defaulting to user installation because normal site-packages is not writeable


Requirement already satisfied: imbalanced-learn in c:\programdata\anaconda3\lib\site-packages (0.10.1)
Requirement already satisfied: scikit-learn>=1.0.2 in c:\programdata\anaconda3\lib\site-packages (from imbalanced-learn) (1.2.1)
Requirement already satisfied: numpy>=1.17.3 in c:\programdata\anaconda3\lib\site-packages (from imbalanced-learn) (1.23.5)
Requirement already satisfied: scipy>=1.3.2 in c:\programdata\anaconda3\lib\site-packages (from imbalanced-learn) (1.10.0)
Requirement already satisfied: threadpoolctl>=2.0.0 in c:\programdata\anaconda3\lib\site-packages (from imbalanced-learn) (2.2.0)
Requirement already satisfied: joblib>=1.1.1 in c:\programdata\anaconda3\lib\site-packages (from imbalanced-learn) (1.1.1)

1 from imblearn.over_sampling import RandomOverSampler


2 ros= RandomOverSampler(random_state=0)
3

1 x_bal,y_bal = ros.fit_resample(x,y)

1 [Link](x=y_bal)

[Link] 5/7
10/29/23, 11:00 PM [Link] - Colaboratory

<Axes: xlabel='Exited', ylabel='count'>

1 from sklearn.model_selection import train_test_split


2 X_train,X_test,Y_train,Y_test = train_test_split(x_bal,y_bal, random_state=2,test_size=0.25)

1 X_train.shape
2

(11944, 9)

1 X_test.shape

(3982, 9)

1 from sklearn.neural_network import MLPClassifier


2 ann=MLPClassifier(hidden_layer_sizes=(100,100,100), max_iter=100,activation="relu", random_state=2)

1 [Link](X_train,Y_train)

▾ MLPClassifier
MLPClassifier(hidden_layer_sizes=(100, 100, 100), max_iter=100, random_state=2)

1 y_pred=[Link](X_test)

1 y_pred
2

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

1 from [Link] import ConfusionMatrixDisplay,accuracy_score,classification_report

1 accuracy_score(Y_test,y_pred)

0.5012556504269211

[Link] 6/7
10/29/23, 11:00 PM [Link] - Colaboratory

[Link] 7/7

Common questions

Powered by AI

The MLPClassifier achieved an accuracy score of 0.8424 on the original test set and 0.5013 on the balanced test set , suggesting that while it initially performed well, balancing the dataset led to poorer performance, indicating potential overfitting on the original dataset.

Overfitting is addressed by implementing cross-validation during the data split in train_test_split, which ensures the model is exposed to different data subsets during training . Additionally, random over-sampling using RandomOverSampler balances the dataset, mitigating overfitting biases . The model's performance is monitored through evaluation metrics like accuracy and F1-score on both original and balanced datasets .

Setting a random state ensures replicability of the results, as it allows for the same random sampling process to be repeated . Using a test size of 0.25 means that 25% of the data is reserved for testing, which allows for a reliable evaluation of the model's performance on unseen data .

The dataset underwent feature scaling using MinMaxScaler and was split into training and test sets using train_test_split. Feature scaling is crucial for normalizing the data, especially when using algorithms like neural networks that are sensitive to the scale of input features . Splitting the data ensures that the model is trained and evaluated on different datasets, which is critical for assessing the model's performance .

The document employed MinMaxScaler for feature scaling to normalize the input features between 0 and 1 . This step is important as it makes the input features suitable for the MLPClassifier by preventing any one feature from dominating the influence on the model due to scale differences, thereby enhancing training convergence and performance .

Using the ReLU activation function can accelerate convergence during training by mitigating the vanishing gradient problem often associated with other activation functions like sigmoid or tanh . ReLU helps the model learn faster and more effectively by allowing for sparse activations, which lead to more efficient computation in the network .

Utilizing multiple hidden layers, as configured with (100, 100, 100), increases the model’s capacity to learn complex patterns by capturing more nuanced features through deeper structures. This enhances the model's ability to generalize patterns from the input data but also requires careful tuning to avoid overfitting, as too many nodes may lead to memorization rather than generalization .

A lower recall score compared to the precision score indicates that the model retrieves fewer true positive instances of class '1'. This suggests that while the model is accurate when it predicts a positive class, it misses predicting many actual positive instances, possibly hinting at sensitivity issues in detecting customers who exited .

Imbalanced datasets can lead to biased models that favor the majority class, reducing accuracy in predicting the minority class. This issue was addressed by employing RandomOverSampler to balance the dataset before retraining the model, which improved the data distribution for more equitable learning .

Precision measures the correctness of positive predictions, recall assesses the ability to find all relevant cases, and F1-score is the harmonic mean of precision and recall, providing a balance between the two . The document used these metrics, alongside accuracy, to evaluate the model's effectiveness, highlighting areas where the model performed well and where it could be improved, especially in terms of class '1' identification .

You might also like