Machine Learning Project: Customer Churn Prediction in Python

import numpy as np
import pandas as pd

rng = np.random.default_rng(7)
n = 4000
contract = rng.choice(["monthly", "yearly"], size=n, p=[0.6, 0.4])
tenure = rng.integers(1, 61, n)                                   # months as a customer
monthly_fee = rng.choice([299, 499, 799], size=n, p=[0.4, 0.4, 0.2])
complaints = rng.poisson(0.6, n)
usage_hours = rng.gamma(4, 5, n).round(1)                         # hours used last month

score = (-1.6 + 1.6 * (contract == "monthly") - 0.04 * tenure + 0.6 * complaints
         + 0.0015 * (monthly_fee - 499) - 0.05 * usage_hours)
churned = rng.random(n) < 1 / (1 + np.exp(-score))

customers = pd.DataFrame({"contract": contract, "tenure": tenure, "monthly_fee": monthly_fee,
                          "complaints": complaints, "usage_hours": usage_hours, "churned": churned.astype(int)})
print(customers.head())
print("churn rate:", round(customers["churned"].mean(), 3))

print(customers.groupby("contract")["churned"].mean().round(3))
print(customers.groupby(pd.cut(customers["tenure"], [0, 6, 24, 60]), observed=True)["churned"].mean().round(3))
print(customers.groupby(customers["complaints"].clip(upper=3))["churned"].mean().round(3))

from sklearn.compose import ColumnTransformer
from sklearn.ensemble import HistGradientBoostingClassifier
from sklearn.linear_model import LogisticRegression
from sklearn.model_selection import cross_validate, train_test_split
from sklearn.pipeline import make_pipeline
from sklearn.preprocessing import OneHotEncoder, StandardScaler

X = customers.drop(columns="churned")
y = customers["churned"]
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.25, random_state=0, stratify=y)

preprocess = ColumnTransformer([
    ("num", StandardScaler(), ["tenure", "monthly_fee", "complaints", "usage_hours"]),
    ("cat", OneHotEncoder(drop="first"), ["contract"]),
])
models = {
    "logistic regression": make_pipeline(preprocess, LogisticRegression(max_iter=1000)),
    "gradient boosting": make_pipeline(preprocess, HistGradientBoostingClassifier(max_depth=3, random_state=0)),
}
print("accuracy of always predicting 'stays':", round(1 - y_train.mean(), 3))
for name, model in models.items():
    scores = cross_validate(model, X_train, y_train, cv=5, scoring=["roc_auc", "average_precision"])
    print(f"{name}: ROC AUC {scores['test_roc_auc'].mean():.3f}, average precision {scores['test_average_precision'].mean():.3f}")

model = models["logistic regression"].fit(X_train, y_train)
names = model[0].get_feature_names_out()
coefs = pd.Series(model[-1].coef_[0], index=names).sort_values()
print(coefs.round(2))

proba = model.predict_proba(X_test)[:, 1]
rows = []
for threshold in [0.1, 0.2, 0.3, 0.4, 0.5, 0.6]:
    contacted = proba >= threshold
    churners_reached = (contacted & (y_test == 1)).sum()
    profit = churners_reached * 0.5 * 4000 - contacted.sum() * 500
    rows.append({"threshold": threshold, "contacted": int(contacted.sum()),
                 "churners_reached": int(churners_reached), "profit_rs": int(profit)})
print(pd.DataFrame(rows).to_string(index=False))

best_threshold = max(rows, key=lambda r: r["profit_rs"])["threshold"]
campaign = X_test.assign(churn_probability=proba.round(2))
campaign = campaign[campaign["churn_probability"] >= best_threshold].sort_values("churn_probability", ascending=False)
print("threshold:", best_threshold, " customers to contact:", len(campaign))
print(campaign.head(5).to_string())