📞 Customer Churn Prediction – Tree Models Mini Project

Build a churn classifier using Decision Tree and Random Forest. Practice preprocessing (encoding, missing values), model training, evaluation, and interpretability (feature importance).

Python: End-to-End Pipeline (Preprocess → Train → Evaluate)

# 1) Imports
import pandas as pd
import numpy as np
from sklearn.model_selection import train_test_split, GridSearchCV
from sklearn.preprocessing import OneHotEncoder
from sklearn.compose import ColumnTransformer
from sklearn.pipeline import Pipeline
from sklearn.metrics import classification_report, confusion_matrix
from sklearn.tree import DecisionTreeClassifier
from sklearn.ensemble import RandomForestClassifier

# 2) Load & basic clean
df = pd.read_csv("churn.csv")  # replace with your path
df = df.dropna(subset=["Churn"])  # ensure target present
df["Churn"] = df["Churn"].map({"Yes":1, "No":0}).astype(int)

# 3) Features/target
y = df["Churn"]
X = df.drop(columns=["Churn","customerID"], errors="ignore")

# 4) Identify column types
num_cols = X.select_dtypes(include=[np.number]).columns.tolist()
cat_cols = X.select_dtypes(exclude=[np.number]).columns.tolist()

# 5) Preprocessor: One-hot for categoricals, pass-through for numerics
preprocess = ColumnTransformer(
    transformers=[
        ("cat", OneHotEncoder(handle_unknown="ignore"), cat_cols)
    ],
    remainder="passthrough"
)

# 6) Models
dt = DecisionTreeClassifier(random_state=42)
rf = RandomForestClassifier(n_estimators=300, random_state=42, n_jobs=-1)

# 7) Pipelines
pipe_dt = Pipeline([("prep", preprocess), ("clf", dt)])
pipe_rf = Pipeline([("prep", preprocess), ("clf", rf)])

# 8) Train/test split (stratify keeps class ratios)
X_tr, X_te, y_tr, y_te = train_test_split(X, y, test_size=0.2, random_state=42, stratify=y)

# 9) Hyperparameter search (quick)
param_dt = {"clf__max_depth":[4,6,8,None], "clf__min_samples_leaf":[1,2,5]}
param_rf = {"clf__max_depth":[None,8,12], "clf__min_samples_leaf":[1,2,4], "clf__max_features":["sqrt","log2"]}

grid_dt = GridSearchCV(pipe_dt, param_dt, cv=5, n_jobs=-1)
grid_rf = GridSearchCV(pipe_rf, param_rf, cv=5, n_jobs=-1)

grid_dt.fit(X_tr, y_tr)
grid_rf.fit(X_tr, y_tr)

best_dt = grid_dt.best_estimator_
best_rf = grid_rf.best_estimator_

# 10) Evaluate
for name, model in [("Decision Tree", best_dt), ("Random Forest", best_rf)]:
    y_pred = model.predict(X_te)
    print(f"\n== {name} ==")
    print(confusion_matrix(y_te, y_pred))
    print(classification_report(y_te, y_pred, digits=3))

Note: For imbalanced churn, monitor Recall (catching churners) and F1. Optionally set class_weight='balanced' in the classifier.

Feature Importance (Random Forest)

# Get feature names after One-Hot
ohe = best_rf.named_steps["prep"].named_transformers_["cat"]
ohe_features = ohe.get_feature_names_out() if len(ohe.categories_) else []
feature_names = list(ohe_features) + num_cols

importances = best_rf.named_steps["clf"].feature_importances_
idx = np.argsort(importances)[-10:][::-1]
print("Top features:")
for i in idx:
    print(feature_names[i], "->", round(importances[i],4))

💡 Improve Your Model

• Add contract length & payment method interactions
• Try class_weight=’balanced’ or resampling (SMOTE) if classes are imbalanced
• Evaluate with AUC-ROC for threshold-independent comparison
• Calibrate probabilities for business decisions (e.g., CalibratedClassifierCV)

📦 What to Submit

• Notebook/script with reproducible pipeline
• Metrics table (Accuracy, Precision, Recall, F1, AUC)
• Top 10 features with brief interpretation
• One actionable recommendation to reduce churn

📝 Self-Check

1. Which metrics matter most for churn and why?
2. How did Random Forest’s feature importance guide your business recommendation?
3. What threshold would you choose for “high churn risk,” and how would you justify it?