Setup

Library Setup

import pandas as pd
from sklearn.model_selection import train_test_split
from sklearn.preprocessing import StandardScaler, OneHotEncoder
from sklearn.compose import ColumnTransformer
from sklearn.pipeline import Pipeline
from sklearn.linear_model import LogisticRegression
from sklearn.metrics import classification_report, confusion_matrix
import joblib

Connect to Google Drive

from google.colab import drive

# Mount Google Drive to save the final model
print("Mounting Google Drive...")
drive.mount('/content/drive')

Mounting Google Drive...
Mounted at /content/drive

Data Setup

import kagglehub

# Download latest version
path = kagglehub.dataset_download("blastchar/telco-customer-churn")

print("Path to dataset files:", path)

# Define the full path to the CSV file
csv_path = "/kaggle/input/telco-customer-churn/WA_Fn-UseC_-Telco-Customer-Churn.csv"

# Load into a DataFrame
data = pd.read_csv(csv_path)

Path to dataset files: /kaggle/input/telco-customer-churn

Data Preparation

Data Exploration

data.head()

	customerID	gender	SeniorCitizen	Partner	Dependents	tenure	PhoneService	MultipleLines	InternetService	OnlineSecurity	...	DeviceProtection	TechSupport	StreamingTV	StreamingMovies	Contract	PaperlessBilling	PaymentMethod	MonthlyCharges	TotalCharges	Churn
0	7590-VHVEG	Female	0	Yes	No	1	No	No phone service	DSL	No	...	No	No	No	No	Month-to-month	Yes	Electronic check	29.85	29.85	No
1	5575-GNVDE	Male	0	No	No	34	Yes	No	DSL	Yes	...	Yes	No	No	No	One year	No	Mailed check	56.95	1889.5	No
2	3668-QPYBK	Male	0	No	No	2	Yes	No	DSL	Yes	...	No	No	No	No	Month-to-month	Yes	Mailed check	53.85	108.15	Yes
3	7795-CFOCW	Male	0	No	No	45	No	No phone service	DSL	Yes	...	Yes	Yes	No	No	One year	No	Bank transfer (automatic)	42.30	1840.75	No
4	9237-HQITU	Female	0	No	No	2	Yes	No	Fiber optic	No	...	No	No	No	No	Month-to-month	Yes	Electronic check	70.70	151.65	Yes

5 rows × 21 columns

data.info()

<class 'pandas.core.frame.DataFrame'>
RangeIndex: 7043 entries, 0 to 7042
Data columns (total 21 columns):
 #   Column            Non-Null Count  Dtype  
---  ------            --------------  -----  
 0   customerID        7043 non-null   object 
 1   gender            7043 non-null   object 
 2   SeniorCitizen     7043 non-null   int64  
 3   Partner           7043 non-null   object 
 4   Dependents        7043 non-null   object 
 5   tenure            7043 non-null   int64  
 6   PhoneService      7043 non-null   object 
 7   MultipleLines     7043 non-null   object 
 8   InternetService   7043 non-null   object 
 9   OnlineSecurity    7043 non-null   object 
 10  OnlineBackup      7043 non-null   object 
 11  DeviceProtection  7043 non-null   object 
 12  TechSupport       7043 non-null   object 
 13  StreamingTV       7043 non-null   object 
 14  StreamingMovies   7043 non-null   object 
 15  Contract          7043 non-null   object 
 16  PaperlessBilling  7043 non-null   object 
 17  PaymentMethod     7043 non-null   object 
 18  MonthlyCharges    7043 non-null   float64
 19  TotalCharges      7043 non-null   object 
 20  Churn             7043 non-null   object 
dtypes: float64(1), int64(2), object(18)
memory usage: 1.1+ MB

We see that all variables should be factors except for tenure, MonthlyCharges, and TotalCharges.

Data Quality

Non-Numeric Values in Numeric Columns

# These columns are numeric
numeric_cols = ['tenure', 'MonthlyCharges', 'TotalCharges']

print("Checking for non-numeric values in expected numeric columns...")
for col in numeric_cols:
    # Try to convert the column to a number
    original_nan_count = data[col].isnull().sum()
    # 'coerce' turns any problems into NaN
    converted_col = pd.to_numeric(data[col], errors='coerce')
    new_nan_count = converted_col.isnull().sum()

    # Check if any new NaNs were created
    if new_nan_count > original_nan_count:
        print(f"  - WARNING: Column '{col}' contains non-numeric values.")
        # Find the problematic rows by comparing the original to the converted column
        problem_rows = data[converted_col.isnull() & data[col].notnull()]
        print(f"    Found {len(problem_rows)} problematic entries. \n\nExample:")
        print(problem_rows.head())
    else:
        print(f"  - OK: Column '{col}' appears to be clean.")

Checking for non-numeric values in expected numeric columns...
  - OK: Column 'tenure' appears to be clean.
  - OK: Column 'MonthlyCharges' appears to be clean.
  - WARNING: Column 'TotalCharges' contains non-numeric values.
    Found 11 problematic entries. 

Example:
      customerID  gender  SeniorCitizen Partner Dependents  tenure  \
488   4472-LVYGI  Female              0     Yes        Yes       0   
753   3115-CZMZD    Male              0      No        Yes       0   
936   5709-LVOEQ  Female              0     Yes        Yes       0   
1082  4367-NUYAO    Male              0     Yes        Yes       0   
1340  1371-DWPAZ  Female              0     Yes        Yes       0   

     PhoneService     MultipleLines InternetService       OnlineSecurity  ...  \
488            No  No phone service             DSL                  Yes  ...   
753           Yes                No              No  No internet service  ...   
936           Yes                No             DSL                  Yes  ...   
1082          Yes               Yes              No  No internet service  ...   
1340           No  No phone service             DSL                  Yes  ...   

         DeviceProtection          TechSupport          StreamingTV  \
488                   Yes                  Yes                  Yes   
753   No internet service  No internet service  No internet service   
936                   Yes                   No                  Yes   
1082  No internet service  No internet service  No internet service   
1340                  Yes                  Yes                  Yes   

          StreamingMovies  Contract PaperlessBilling  \
488                    No  Two year              Yes   
753   No internet service  Two year               No   
936                   Yes  Two year               No   
1082  No internet service  Two year               No   
1340                   No  Two year               No   

                  PaymentMethod MonthlyCharges  TotalCharges Churn  
488   Bank transfer (automatic)          52.55                  No  
753                Mailed check          20.25                  No  
936                Mailed check          80.85                  No  
1082               Mailed check          25.75                  No  
1340    Credit card (automatic)          56.05                  No  

[5 rows x 21 columns]

Missing Value Summary

# Clean 'TotalCharges'
print("Performing final check for missing values after cleaning...")
data['TotalCharges'] = pd.to_numeric(data['TotalCharges'], errors='coerce')

missing_values = data.isnull().sum()
# Filter to only show columns with missing data
missing_values = missing_values[missing_values > 0]

# Do a final check for any NaNs
if missing_values.empty:
    print("  - SUCCESS: No missing values found in any columns after cleaning.")
else:
    print("  - WARNING: Missing values found in the following columns:")
    print(missing_values)

# Drop rows with any missing values
data.dropna(inplace=True)

Performing final check for missing values after cleaning...
  - WARNING: Missing values found in the following columns:
TotalCharges    11
dtype: int64

There were only 11 rows where TotalCharges was missing, so I drop them.

Unique Values in Categorical Columns

# Get a list of all columns of type 'object'
categorical_cols = data.select_dtypes(include=['object']).columns

print("\n[INFO] Inspecting unique values in categorical columns...")
for col in categorical_cols:
    # Ignore 'customerID' as its values are all unique
    if col != 'customerID':
        unique_values = data[col].unique()
        print(f"  - Column '{col}' has {len(unique_values)} unique values: {unique_values}")

[INFO] Inspecting unique values in categorical columns...
  - Column 'gender' has 2 unique values: ['Female' 'Male']
  - Column 'Partner' has 2 unique values: ['Yes' 'No']
  - Column 'Dependents' has 2 unique values: ['No' 'Yes']
  - Column 'PhoneService' has 2 unique values: ['No' 'Yes']
  - Column 'MultipleLines' has 3 unique values: ['No phone service' 'No' 'Yes']
  - Column 'InternetService' has 3 unique values: ['DSL' 'Fiber optic' 'No']
  - Column 'OnlineSecurity' has 3 unique values: ['No' 'Yes' 'No internet service']
  - Column 'OnlineBackup' has 3 unique values: ['Yes' 'No' 'No internet service']
  - Column 'DeviceProtection' has 3 unique values: ['No' 'Yes' 'No internet service']
  - Column 'TechSupport' has 3 unique values: ['No' 'Yes' 'No internet service']
  - Column 'StreamingTV' has 3 unique values: ['No' 'Yes' 'No internet service']
  - Column 'StreamingMovies' has 3 unique values: ['No' 'Yes' 'No internet service']
  - Column 'Contract' has 3 unique values: ['Month-to-month' 'One year' 'Two year']
  - Column 'PaperlessBilling' has 2 unique values: ['Yes' 'No']
  - Column 'PaymentMethod' has 4 unique values: ['Electronic check' 'Mailed check' 'Bank transfer (automatic)'
 'Credit card (automatic)']
  - Column 'Churn' has 2 unique values: ['No' 'Yes']

Data Preprocessing

Feature Encoding

# Drop 'customerID' as it's just an identifier
data.drop('customerID', axis=1, inplace=True)

# Convert target 'Churn' to (0, 1)
data['Churn'] = data['Churn'].apply(lambda x: 1 if x == 'Yes' else 0)

# Separate the features from the target
X = data.drop('Churn', axis=1)
y = data['Churn']

# Separate categorical and numerical features
categorical_features = X.select_dtypes(include=['object']).columns
numerical_features = X.select_dtypes(include=['int64', 'float64']).columns

# Create a preprocessing pipeline
preprocessor = ColumnTransformer(
    transformers=[
        # Scale the numerical features
        ('num', StandardScaler(), numerical_features),
        # One-hot encode the categorical features
        ('cat', OneHotEncoder(handle_unknown='ignore'), categorical_features)
    ])

# Split the data into training and testing sets
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.2, random_state=42, stratify=y)

Building a Model

I choose to use logistic regression, as it is a simple but powerful model for classification.

Define Model

# Create the full model pipeline
model = Pipeline(steps=[('preprocessor', preprocessor),
                        ('classifier', LogisticRegression(random_state=42, max_iter=1000))])

Training Model

model.fit(X_train, y_train)

Pipeline(steps=[('preprocessor',
                 ColumnTransformer(transformers=[('num', StandardScaler(),
                                                  Index(['SeniorCitizen', 'tenure', 'MonthlyCharges', 'TotalCharges'], dtype='object')),
                                                 ('cat',
                                                  OneHotEncoder(handle_unknown='ignore'),
                                                  Index(['gender', 'Partner', 'Dependents', 'PhoneService', 'MultipleLines',
       'InternetService', 'OnlineSecurity', 'OnlineBackup', 'DeviceProtection',
       'TechSupport', 'StreamingTV', 'StreamingMovies', 'Contract',
       'PaperlessBilling', 'PaymentMethod'],
      dtype='object'))])),
                ('classifier',
                 LogisticRegression(max_iter=1000, random_state=42))])

In a Jupyter environment, please rerun this cell to show the HTML representation or trust the notebook.
On GitHub, the HTML representation is unable to render, please try loading this page with nbviewer.org.

Model Evaluation

# Make predictions on the test set
y_pred = model.predict(X_test)

# Print a classification report
print("Classification Report:")
print(classification_report(y_test, y_pred))

# Print the confusion matrix
print("Confusion Matrix:")
print(confusion_matrix(y_test, y_pred))

Classification Report:
              precision    recall  f1-score   support

           0       0.85      0.89      0.87      1033
           1       0.65      0.57      0.61       374

    accuracy                           0.80      1407
   macro avg       0.75      0.73      0.74      1407
weighted avg       0.80      0.80      0.80      1407

Confusion Matrix:
[[917 116]
 [160 214]]

Feature Importance

# Extract the trained classifier and preprocessor from the pipeline
final_classifier = model.named_steps['classifier']
final_preprocessor = model.named_steps['preprocessor']

# Get the feature names after one-hot encoding
ohe_feature_names = final_preprocessor.named_transformers_['cat'].get_feature_names_out(categorical_features)
all_feature_names = list(numerical_features) + list(ohe_feature_names)

# Get the coefficients from the model
coefficients = final_classifier.coef_[0]

# Create a DataFrame to view the feature importances
feature_importance = pd.DataFrame({'Feature': all_feature_names, 'Importance': coefficients})
feature_importance = feature_importance.sort_values(by='Importance', ascending=False)

print("\nTop 10 Features Increasing Churn Likelihood:")
print(feature_importance.head(10))

print("\nTop 10 Features Decreasing Churn Likelihood (Promoting Retention):")
print(feature_importance.tail(10))

Top 10 Features Increasing Churn Likelihood:
                           Feature  Importance
3                     TotalCharges    0.644014
36         Contract_Month-to-month    0.613846
16     InternetService_Fiber optic    0.590184
32                 StreamingTV_Yes    0.191113
43  PaymentMethod_Electronic check    0.180671
35             StreamingMovies_Yes    0.177076
18               OnlineSecurity_No    0.164459
27                  TechSupport_No    0.142949
14               MultipleLines_Yes    0.088164
0                    SeniorCitizen    0.071011

Top 10 Features Decreasing Churn Likelihood (Promoting Retention):
                                 Feature  Importance
19    OnlineSecurity_No internet service   -0.283724
17                    InternetService_No   -0.283724
31       StreamingTV_No internet service   -0.283724
25  DeviceProtection_No internet service   -0.283724
12                      MultipleLines_No   -0.293470
39                   PaperlessBilling_No   -0.300387
2                         MonthlyCharges   -0.541006
15                   InternetService_DSL   -0.616132
38                     Contract_Two year   -0.779030
1                                 tenure   -1.352313

import seaborn as sns
import matplotlib.pyplot as plt

# Visualize feature importance
plt.style.use('seaborn-v0_8-whitegrid')
top_and_bottom_10 = pd.concat([feature_importance.head(10), feature_importance.tail(10)])

# Add a dummy hue column to use with palette (fixes sns warning)
top_and_bottom_10['dummy_hue'] = top_and_bottom_10['Importance']

plt.figure(figsize=(12, 10))
sns.barplot(
    x='Importance',
    y='Feature',
    data=top_and_bottom_10,
    hue='dummy_hue',
    palette='coolwarm',
    dodge=False,
    legend=False
)
plt.title('Top 20 Most Important Features for Predicting Churn', fontsize=16)
plt.xlabel('Coefficient (Importance)', fontsize=12)
plt.ylabel('Feature', fontsize=12)
plt.tight_layout()
plt.show()

Least Important Features

I calculate the least important features to potentially prune them and simplify the model.

import numpy as np
# Calculate the absolute importance to find features closest to zero
feature_importance['Absolute_Importance'] = np.abs(feature_importance['Importance'])

# Sort by the absolute importance in ascending order
least_important_features = feature_importance.sort_values(by='Absolute_Importance', ascending=True)

print("Features ranked by their impact (least important first):")
# Display all features, starting with the least important
print(least_important_features.to_string())

Features ranked by their impact (least important first):
                                    Feature  Importance  Absolute_Importance
40                     PaperlessBilling_Yes   -0.009285             0.009285
26                     DeviceProtection_Yes    0.032693             0.032693
8                             Dependents_No   -0.039151             0.039151
21                          OnlineBackup_No    0.043639             0.043639
24                      DeviceProtection_No   -0.058642             0.058642
23                         OnlineBackup_Yes   -0.069587             0.069587
0                             SeniorCitizen    0.071011             0.071011
14                        MultipleLines_Yes    0.088164             0.088164
13           MultipleLines_No phone service   -0.104365             0.104365
10                          PhoneService_No   -0.104365             0.104365
44               PaymentMethod_Mailed check   -0.123531             0.123531
6                                Partner_No   -0.142419             0.142419
4                             gender_Female   -0.142779             0.142779
27                           TechSupport_No    0.142949             0.142949
37                        Contract_One year   -0.144488             0.144488
42    PaymentMethod_Credit card (automatic)   -0.162056             0.162056
18                        OnlineSecurity_No    0.164459             0.164459
5                               gender_Male   -0.166893             0.166893
7                               Partner_Yes   -0.167253             0.167253
29                          TechSupport_Yes   -0.168897             0.168897
35                      StreamingMovies_Yes    0.177076             0.177076
43           PaymentMethod_Electronic check    0.180671             0.180671
20                       OnlineSecurity_Yes   -0.190408             0.190408
32                          StreamingTV_Yes    0.191113             0.191113
33                       StreamingMovies_No   -0.203025             0.203025
41  PaymentMethod_Bank transfer (automatic)   -0.204755             0.204755
11                         PhoneService_Yes   -0.205307             0.205307
30                           StreamingTV_No   -0.217061             0.217061
9                            Dependents_Yes   -0.270520             0.270520
28          TechSupport_No internet service   -0.283724             0.283724
34      StreamingMovies_No internet service   -0.283724             0.283724
22         OnlineBackup_No internet service   -0.283724             0.283724
17                       InternetService_No   -0.283724             0.283724
31          StreamingTV_No internet service   -0.283724             0.283724
25     DeviceProtection_No internet service   -0.283724             0.283724
19       OnlineSecurity_No internet service   -0.283724             0.283724
12                         MultipleLines_No   -0.293470             0.293470
39                      PaperlessBilling_No   -0.300387             0.300387
2                            MonthlyCharges   -0.541006             0.541006
16              InternetService_Fiber optic    0.590184             0.590184
36                  Contract_Month-to-month    0.613846             0.613846
15                      InternetService_DSL   -0.616132             0.616132
3                              TotalCharges    0.644014             0.644014
38                        Contract_Two year   -0.779030             0.779030
1                                    tenure   -1.352313             1.352313

Pruned Model

# Create a copy to avoid modifying the original DataFrame
data_pruned = data.copy()

# 1. Prune the unimportant features
features_to_drop = ['gender', 'PhoneService', 'Partner']
data_pruned.drop(columns=features_to_drop, inplace=True)
print(f"Dropped columns: {features_to_drop}\n")

# 2. Set up the new feature set and target variable
X_pruned = data_pruned.drop('Churn', axis=1)
y_pruned = data_pruned['Churn']

# Identify column types for the new, smaller dataset
categorical_features_pruned = X_pruned.select_dtypes(include=['object']).columns
numerical_features_pruned = X_pruned.select_dtypes(include=['int64', 'float64']).columns

# 3. Split the pruned data
X_train_p, X_test_p, y_train_p, y_test_p = train_test_split(X_pruned, y_pruned, test_size=0.2, random_state=42, stratify=y_pruned)

# 4. Define a new preprocessor and pipeline for the pruned data
preprocessor_pruned = ColumnTransformer(
    transformers=[
        ('num', StandardScaler(), numerical_features_pruned),
        ('cat', OneHotEncoder(handle_unknown='ignore'), categorical_features_pruned)
    ])

model_pruned = Pipeline(steps=[('preprocessor', preprocessor_pruned),
                               ('classifier', LogisticRegression(random_state=42, max_iter=1000))])

# 5. Train the new model
model_pruned.fit(X_train_p, y_train_p)

# 6. Evaluate the new model's performance
print("Classification Report:")
y_pred_p = model_pruned.predict(X_test_p)
print(classification_report(y_test_p, y_pred_p))

Dropped columns: ['gender', 'PhoneService', 'Partner']

Classification Report:
              precision    recall  f1-score   support

           0       0.85      0.89      0.87      1033
           1       0.65      0.57      0.61       374

    accuracy                           0.80      1407
   macro avg       0.75      0.73      0.74      1407
weighted avg       0.80      0.80      0.80      1407

Save Model

# Save the trained pruned model
model_name = 'churn_model.joblib'
drive_path = f'/content/drive/My Drive/{model_name}'
# Uncomment this if you want the full model
#joblib.dump(model, drive_path)
joblib.dump(model_pruned, drive_path)
print(f"Best model saved to Google Drive at: {drive_path}")

Best model saved to Google Drive at: /content/drive/My Drive/churn_model.joblib