Telecom Customer Churn Prediction

A Model that will predict whether a customer will churn or not based on some given inputs (features)

1. Data Loading

Here I am loading data from a csv file through Pandas library.

# Import some basic libraries for data science
import pandas as pd
import numpy as np
import seaborn as sb
import matplotlib.pyplot as ml
from sklearn.model_selection import train_test_split

# Loading the CSV dataset into code
customer_df = pd.read_csv("/Users/haseebsagheer/Documents/Python Learning/Customer-Churn/Datasets/WA_Fn-UseC_-Telco-Customer-Churn 3.csv")
# Setting the view limit of columns to max
pd.set_option('display.max_columns', None)  # Show all columns
pd.set_option('display.expand_frame_repr', False)  # Prevent line wrapping

# Need to check the header values
print(customer_df.head())

   customerID  gender  SeniorCitizen Partner Dependents  tenure PhoneService     MultipleLines InternetService OnlineSecurity OnlineBackup DeviceProtection TechSupport StreamingTV StreamingMovies        Contract PaperlessBilling              PaymentMethod  MonthlyCharges TotalCharges Churn
0  7590-VHVEG  Female              0     Yes         No       1           No  No phone service             DSL             No          Yes               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           No              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          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           No              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              No  Month-to-month              Yes           Electronic check           70.70       151.65   Yes

2. Data Understanding

Understanding the data through some useful funtions in Python, so i can get a quick overview of data such as describe(), info() etc.

customer_df.shape

(7043, 21)

customer_df.columns

Index(['customerID', 'gender', 'SeniorCitizen', 'Partner', 'Dependents',
       'tenure', 'PhoneService', 'MultipleLines', 'InternetService',
       'OnlineSecurity', 'OnlineBackup', 'DeviceProtection', 'TechSupport',
       'StreamingTV', 'StreamingMovies', 'Contract', 'PaperlessBilling',
       'PaymentMethod', 'MonthlyCharges', 'TotalCharges', 'Churn'],
      dtype='object')

customer_df.head()

	customerID	gender	SeniorCitizen	Partner	Dependents	tenure	PhoneService	MultipleLines	InternetService	OnlineSecurity	OnlineBackup	DeviceProtection	TechSupport	StreamingTV	StreamingMovies	Contract	PaperlessBilling	PaymentMethod	MonthlyCharges	TotalCharges	Churn
0	7590-VHVEG	Female	0	Yes	No	1	No	No phone service	DSL	No	Yes	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	No	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	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	No	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	No	Month-to-month	Yes	Electronic check	70.70	151.65	Yes

customer_df.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

customer_df.describe()

	SeniorCitizen	tenure	MonthlyCharges
count	7043.000000	7043.000000	7043.000000
mean	0.162147	32.371149	64.761692
std	0.368612	24.559481	30.090047
min	0.000000	0.000000	18.250000
25%	0.000000	9.000000	35.500000
50%	0.000000	29.000000	70.350000
75%	0.000000	55.000000	89.850000
max	1.000000	72.000000	118.750000

2.1 Initial Data Visualization

Doing some intial visualization between some features and churn

sb.countplot(x="Churn", data=customer_df)
ml.title("Customer Churn Distribution")
ml.xlabel("Churn (Yes / No)")
ml.ylabel("Number of Customers")
ml.show()

70% of the users have not churned yet, and 30% have already done so.

sb.countplot(data=customer_df, x="gender", hue="Churn")
ml.title("Churn by Gender")
ml.xlabel("Gender")
ml.ylabel("Count")
ml.show()

Outcome

1. Customer Churn is not significantly impacted by gender. The number of female customers who churned is very similar to the number of male customers who churned.
2. The majority of customers, for both genders, have not churned. The blue bars for both females and males are significantly taller than their corresponding orange bars.
3. The total number of male and female customers is nearly equal. The combined height of the "No" and "Yes" bars for females is very close to the combined height for males.

sb.countplot(data=customer_df,x="Contract",hue="Churn")
ml.title("Churn by Contract Type")

Text(0.5, 1.0, 'Churn by Contract Type')

Contract type is a strong predictor of churn

sb.boxplot(data=customer_df, x="Churn", y="tenure")
ml.title("Tenure Distribution by Churn")
ml.xlabel("Churn")
ml.ylabel("Tenure")
ml.show()

sb.barplot(data=customer_df, x="Churn", y="tenure", estimator="mean")
ml.title("Average Tenure by Churn")
ml.xlabel("Churn")
ml.ylabel("Average Tenure")
ml.show()

customer_df.head()

	customerID	gender	SeniorCitizen	Partner	Dependents	tenure	PhoneService	MultipleLines	InternetService	OnlineSecurity	OnlineBackup	DeviceProtection	TechSupport	StreamingTV	StreamingMovies	Contract	PaperlessBilling	PaymentMethod	MonthlyCharges	TotalCharges	Churn
0	7590-VHVEG	Female	0	Yes	No	1	No	No phone service	DSL	No	Yes	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	No	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	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	No	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	No	Month-to-month	Yes	Electronic check	70.70	151.65	Yes

3. Data Preprocessing (Missing Values, Removing Extra Feature etc)

customer_df.drop("customerID",axis= 1,inplace=True)
customer_df.drop("gender",axis = 1, inplace=True)

customer_df["Partner"].value_counts()
def binary_encoding(column):
    if column.nunique() != 2:
        print("Unable to convert to binary")
        return column
    mapping = {"Yes":1,"No":0}
    return column.replace(mapping).astype(int)

customer_df["Partner"] = binary_encoding(customer_df["Partner"])
customer_df["Dependents"] = binary_encoding(customer_df["Dependents"])
customer_df["PhoneService"] = binary_encoding(customer_df["PhoneService"])
customer_df["PaperlessBilling"] = binary_encoding(customer_df["PaperlessBilling"])
customer_df["Churn"] = binary_encoding(customer_df["Churn"])

/var/folders/8n/ybp5gv0n715b8994769t9_f00000gn/T/ipykernel_15552/3136517368.py:7: FutureWarning: Downcasting behavior in `replace` is deprecated and will be removed in a future version. To retain the old behavior, explicitly call `result.infer_objects(copy=False)`. To opt-in to the future behavior, set `pd.set_option('future.no_silent_downcasting', True)`
  return column.replace(mapping).astype(int)
/var/folders/8n/ybp5gv0n715b8994769t9_f00000gn/T/ipykernel_15552/3136517368.py:7: FutureWarning: Downcasting behavior in `replace` is deprecated and will be removed in a future version. To retain the old behavior, explicitly call `result.infer_objects(copy=False)`. To opt-in to the future behavior, set `pd.set_option('future.no_silent_downcasting', True)`
  return column.replace(mapping).astype(int)
/var/folders/8n/ybp5gv0n715b8994769t9_f00000gn/T/ipykernel_15552/3136517368.py:7: FutureWarning: Downcasting behavior in `replace` is deprecated and will be removed in a future version. To retain the old behavior, explicitly call `result.infer_objects(copy=False)`. To opt-in to the future behavior, set `pd.set_option('future.no_silent_downcasting', True)`
  return column.replace(mapping).astype(int)
/var/folders/8n/ybp5gv0n715b8994769t9_f00000gn/T/ipykernel_15552/3136517368.py:7: FutureWarning: Downcasting behavior in `replace` is deprecated and will be removed in a future version. To retain the old behavior, explicitly call `result.infer_objects(copy=False)`. To opt-in to the future behavior, set `pd.set_option('future.no_silent_downcasting', True)`
  return column.replace(mapping).astype(int)
/var/folders/8n/ybp5gv0n715b8994769t9_f00000gn/T/ipykernel_15552/3136517368.py:7: FutureWarning: Downcasting behavior in `replace` is deprecated and will be removed in a future version. To retain the old behavior, explicitly call `result.infer_objects(copy=False)`. To opt-in to the future behavior, set `pd.set_option('future.no_silent_downcasting', True)`
  return column.replace(mapping).astype(int)

customer_df= pd.get_dummies(data=customer_df,columns=["MultipleLines"],drop_first=True)
customer_df = pd.get_dummies(data=customer_df,columns=["InternetService"],drop_first=True)
customer_df = pd.get_dummies(data=customer_df,columns=["OnlineSecurity"],drop_first=True)
customer_df = pd.get_dummies(data=customer_df,columns=["OnlineBackup"],drop_first=True)
customer_df = pd.get_dummies(data=customer_df,columns=["DeviceProtection"],drop_first=True)
customer_df = pd.get_dummies(data=customer_df,columns=["TechSupport"],drop_first=True)
customer_df = pd.get_dummies(data=customer_df,columns=["StreamingTV"],drop_first=True)
customer_df = pd.get_dummies(data=customer_df,columns=["StreamingMovies"],drop_first=True)
customer_df = pd.get_dummies(data=customer_df,columns=["Contract"],drop_first=True)
customer_df = pd.get_dummies(data=customer_df,columns=["PaperlessBilling"],drop_first=True)
customer_df = pd.get_dummies(data=customer_df,columns=["PaymentMethod"],drop_first=True)
customer_df.head()

	SeniorCitizen	Partner	Dependents	tenure	PhoneService	MonthlyCharges	TotalCharges	Churn	MultipleLines_No phone service	MultipleLines_Yes	InternetService_Fiber optic	InternetService_No	OnlineSecurity_No internet service	OnlineSecurity_Yes	OnlineBackup_No internet service	OnlineBackup_Yes	DeviceProtection_No internet service	DeviceProtection_Yes	TechSupport_No internet service	TechSupport_Yes	StreamingTV_No internet service	StreamingTV_Yes	StreamingMovies_No internet service	StreamingMovies_Yes	Contract_One year	Contract_Two year	PaperlessBilling_1	PaymentMethod_Credit card (automatic)	PaymentMethod_Electronic check	PaymentMethod_Mailed check
0	0	1	0	1	0	29.85	29.85	0	True	False	False	False	False	False	False	True	False	False	False	False	False	False	False	False	False	False	True	False	True	False
1	0	0	0	34	1	56.95	1889.5	0	False	False	False	False	False	True	False	False	False	True	False	False	False	False	False	False	True	False	False	False	False	True
2	0	0	0	2	1	53.85	108.15	1	False	False	False	False	False	True	False	True	False	False	False	False	False	False	False	False	False	False	True	False	False	True
3	0	0	0	45	0	42.30	1840.75	0	True	False	False	False	False	True	False	False	False	True	False	True	False	False	False	False	True	False	False	False	False	False
4	0	0	0	2	1	70.70	151.65	1	False	False	True	False	False	False	False	False	False	False	False	False	False	False	False	False	False	False	True	False	True	False

pd.set_option("display.width",None)
pd.set_option("display.max_rows",None)
pd.set_option("display.max_row",None)
customer_df = customer_df[pd.to_numeric(customer_df["TotalCharges"], errors="coerce").notnull()]
customer_df["TotalCharges"] = pd.to_numeric(customer_df["TotalCharges"], errors="coerce")

4. Modeling Prepration

X = customer_df.drop("Churn",axis = 1)
Y = customer_df["Churn"]

X_train, X_test,Y_train,Y_test = train_test_split(X,Y,test_size=20,stratify=Y,random_state=42)

print(Y.value_counts(normalize=True))        # original distribution
print(Y_train.value_counts(normalize=True))  # training set
print(Y_test.value_counts(normalize=True))   # test set

Churn
0    0.734215
1    0.265785
Name: proportion, dtype: float64
Churn
0    0.73417
1    0.26583
Name: proportion, dtype: float64
Churn
0    0.75
1    0.25
Name: proportion, dtype: float64

5. Modeling Training

from sklearn.linear_model import LogisticRegression
from sklearn.metrics import accuracy_score, confusion_matrix, classification_report, roc_auc_score, roc_curve

model = LogisticRegression(max_iter=1000, solver='liblinear')  # You can try 'lbfgs' too
model.fit(X_train, Y_train)
y_pred = model.predict(X_test)                       # 0 or 1
y_proba = model.predict_proba(X_test)[:, 1]          # Probability of class = 1 (Churn)
# Accuracy
print("Accuracy:", accuracy_score(y_test, y_pred))

# Classification report
print("\nClassification Report:\n", classification_report(y_test, y_pred))

# Confusion Matrix
conf_matrix = confusion_matrix(y_test, y_pred)
sb.heatmap(conf_matrix, annot=True, fmt='d', cmap='Blues')
ml.title("Confusion Matrix")
ml.xlabel("Predicted")
ml.ylabel("Actual")
ml.show()

Accuracy: 0.8

Classification Report:
               precision    recall  f1-score   support

           0       0.92      0.80      0.86        15
           1       0.57      0.80      0.67         5

    accuracy                           0.80        20
   macro avg       0.75      0.80      0.76        20
weighted avg       0.84      0.80      0.81        20

fpr, tpr, thresholds = roc_curve(Y_test, y_proba)
roc_auc = roc_auc_score(Y_test, y_proba)

ml.figure(figsize=(6, 4))
ml.plot(fpr, tpr, label=f"ROC Curve (AUC = {roc_auc:.2f})")
ml.plot([0, 1], [0, 1], linestyle="--", color="gray")
ml.xlabel("False Positive Rate")
ml.ylabel("True Positive Rate")
ml.title("Receiver Operating Characteristic (ROC) Curve")
ml.legend(loc="lower right")
ml.grid()

ml.savefig("roc_curve.png", dpi=300, bbox_inches='tight') 
ml.show()

📊 Model Evaluation Summary (Logistic Regression)

Our baseline Logistic Regression model achieved strong performance. Here's a summary of its evaluation:

✅ Accuracy

• Accuracy: 0.80
• This means the model correctly predicted churn or no-churn for 80% of the customers.

✅ Classification Report

Metric	Class 0 (Not Churned)	Class 1 (Churned)
Precision	0.92	0.57
Recall	0.80	0.80
F1-score	0.86	0.67
Support	15	5

Interpretation:

• The model is very good at correctly identifying customers who won’t churn (class 0).
• It correctly identifies 80% of actual churners (high recall), but with some false positives (lower precision for class 1).
• Overall, F1-score of 0.67 for churners is a solid starting point.

---

📈 ROC Curve and AUC Score

The ROC (Receiver Operating Characteristic) curve helps us understand how well the model separates churn vs. non-churn across all thresholds.

• AUC (Area Under Curve): 0.84
• This means there's an 84% chance the model ranks a random churner higher than a non-churner.

Conclusion:
The steep rise in the ROC curve and the high AUC score demonstrate that our model is capable of distinguishing between churn and non-churn customers effectively.

---

🌲 Random Forest Classifier - Model Training

Random Forest is an ensemble method that builds multiple decision trees and combines their outputs to improve predictive performance and reduce overfitting.

In this step, we:

• Trained a Random Forest model using the processed churn dataset
• Used X_train and y_train to fit the model
• Will evaluate performance on X_test and y_test using accuracy, precision, recall, F1-score, and AUC

Benefits of Random Forest:

• Handles non-linear relationships well
• Automatically manages feature interactions
• Less sensitive to outliers and noise

🔧 Random Forest Tuning - First Attempt

We tuned the following hyperparameters:

• n_estimators=200: More trees for better generalization
• max_depth=8: Prevents overly deep trees and overfitting
• min_samples_split=5 and min_samples_leaf=2: Smoother splits
• class_weight='balanced': Handles class imbalance for better recall

This configuration aims to improve prediction of minority class (churners).

from sklearn.ensemble import RandomForestClassifier
# 1. Initialize the model
rf_model = RandomForestClassifier(
    n_estimators=200,
    max_depth=8,
    min_samples_split=5,
    min_samples_leaf=2,
    class_weight='balanced',
    random_state=42
)


# 2. Train the model
rf_model.fit(X_train, Y_train)

# 3. Predict
y_pred = rf_model.predict(X_test)
y_prob = rf_model.predict_proba(X_test)[:, 1]  # for ROC AUC

# 4. Evaluate
print("Confusion Matrix:")
print(confusion_matrix(Y_test, y_pred))
print("\nClassification Report:")
print(classification_report(Y_test, y_pred))

# 5. ROC Curve
fpr, tpr, _ = roc_curve(Y_test, y_prob)
auc_score = roc_auc_score(Y_test, y_prob)

ml.plot(fpr, tpr, label=f"Random Forest (AUC = {auc_score:.2f})")
ml.plot([0, 1], [0, 1], linestyle="--", color="gray")
ml.xlabel("False Positive Rate")
ml.ylabel("True Positive Rate")
ml.title("ROC Curve - Random Forest Classifier")
ml.legend()
ml.grid()
ml.savefig("random_forest_roc.png", dpi=300, bbox_inches='tight')
ml.show()

Confusion Matrix:
[[11  4]
 [ 0  5]]

Classification Report:
              precision    recall  f1-score   support

           0       1.00      0.73      0.85        15
           1       0.56      1.00      0.71         5

    accuracy                           0.80        20
   macro avg       0.78      0.87      0.78        20
weighted avg       0.89      0.80      0.81        20

# 1. Extract feature importances
importances = rf_model.feature_importances_

# 2. Match them to column names
features = X_train.columns

# 3. Create a DataFrame for easy sorting and plotting
importance_df = pd.DataFrame({
    'Feature': features,
    'Importance': importances
}).sort_values(by="Importance", ascending=True)

# 4. Plot it
import matplotlib.pyplot as plt

ml.figure(figsize=(10, 8))
ml.barh(importance_df['Feature'], importance_df['Importance'])
ml.xlabel("Feature Importance Score")
ml.title("Random Forest Feature Importance")
ml.grid(True)
ml.tight_layout()
ml.savefig("rf_feature_importance.png", dpi=300)
ml.show()

🔍 Feature Importance – Random Forest

We used Random Forest's built-in feature importance to identify the top drivers of churn:

• tenure: Most powerful predictor — new customers are more likely to leave
• TotalCharges: Low-spending customers tend to churn
• Contract_Two year: Long-term contracts reduce churn probability
• InternetService_Fiber optic: May be linked with dissatisfaction or cost
• MonthlyCharges: High bills could trigger churn

Features with very low importance (like PhoneService, MultipleLines, StreamingTV_Yes) had minimal predictive power.

These insights help the business understand customer behavior and can guide retention strategies.

📦 Project Wrap-Up: Telco Customer Churn Prediction

After spending considerable time on this project, I’m wrapping it up with a deep understanding of not just the model outputs, but also how and why customers churn. This was more than just a machine learning exercise — it was a complete data science lifecycle experience, from raw data to insights and action.

---

✅ What I Did

🔹 I started by exploring the Telco Customer Churn dataset from Kaggle.
🔹 The dataset had no missing/null values — however, TotalCharges had some values as strings, so I cleaned and converted it to numeric.

🔹 I then did extensive preprocessing:

• Binary encoded Yes/No columns
• Applied One-Hot Encoding to multi-class categorical features
• Dropped irrelevant features like customerID and gender based on insight and balance

🔹 I visualized churn distribution using Seaborn:

• Compared churn by gender, contract type, payment method, and more
• Learned valuable business-level insights from visualizations before modeling

---

🧠 Modeling Journey

I trained two models:

1. Logistic Regression:

• Simple, interpretable, and great as a baseline
• Achieved 80% accuracy, with decent precision and recall
• Created an ROC curve to evaluate classification performance (AUC = 0.84)

2. Random Forest Classifier:

• A more complex ensemble model
• After tuning, also achieved 80% accuracy
• AUC score slightly improved to 0.84
• Gave me a ranked list of feature importances, which was a goldmine of insight

---

💡 Key Takeaways

• Tenure was the top predictor of churn: newer customers are more likely to leave.
• Customers on long-term contracts tend to stay.
• High monthly charges and low total charges indicate a higher churn risk.
• Not all features matter: I found that columns like PhoneService, MultipleLines, and some streaming service flags had minimal impact.

---

📊 Final Thoughts

This wasn’t just about model accuracy — it was about understanding why customers leave, what drives loyalty, and how a company could act on this data.

I now feel confident in:

• End-to-end data preparation
• Feature encoding strategies
• Model selection and evaluation
• Visual storytelling for business stakeholders

---

I plan to revisit this project later and possibly:

• Add SHAP explanations for per-customer churn drivers
• Integrate with a web dashboard for interactive insights
• Try XGBoost or Neural Networks for further optimization

Thanks for following my journey — and more projects are on the way! 🚀

import nbformat
from nbconvert import HTMLExporter
from traitlets.config import Config

nb = nbformat.read("customer_churn.ipynb", as_version=4)

c = Config()
# Keep code + markdown + outputs
c.HTMLExporter.exclude_input = False
c.HTMLExporter.exclude_output = False

exporter = HTMLExporter(config=c)
body, _ = exporter.from_notebook_node(nb)

with open("customer_churn.html", "w", encoding="utf-8") as f:
    f.write(body)

print("Done -> YourNotebook.html")