दिन 56: Project 1: Business Problem और Data

import pandas as pd
raw = pd.read_csv("telecom_churn.csv")
print(raw.shape)
print(raw.dtypes.astype(str).to_dict())
print("Missing values:\n", raw.isna().sum()[raw.isna().sum() > 0])
print("internet_service values:", raw["internet_service"].value_counts(dropna=False).to_dict())

import numpy as np
import pandas as pd

def load_churn(path="telecom_churn.csv"):
    # "None" means no internet plan – stop pandas from turning it into NaN
    df = pd.read_csv(path, keep_default_na=False, na_values=[""])
    df["total_charges"] = df["total_charges"].fillna(df["monthly_charges"] * df["tenure_months"])
    df["churned"] = (df["churn"] == "Yes").astype(int)
    return df

df = load_churn()
print(df["internet_service"].value_counts().to_dict())
print("Missing after load:", int(df.isna().sum().sum()))
print("Duplicates:", df["customer_id"].duplicated().sum())
check = (df["tech_support"] == "No internet") == (df["internet_service"] == "None")
print("tech_support consistent with internet_service:", bool(check.all()))

import numpy as np
import pandas as pd

def load_churn(path="telecom_churn.csv"):
    # "None" means no internet plan – stop pandas from turning it into NaN
    df = pd.read_csv(path, keep_default_na=False, na_values=[""])
    df["total_charges"] = df["total_charges"].fillna(df["monthly_charges"] * df["tenure_months"])
    df["churned"] = (df["churn"] == "Yes").astype(int)
    return df

df = load_churn()
rules = {"tenure 1–72 months": df["tenure_months"].between(1, 72).all(),
         "monthly charges > 0": (df["monthly_charges"] > 0).all(),
         "total ≈ monthly × tenure": (abs(df["total_charges"] - df["monthly_charges"] * df["tenure_months"]) / df["total_charges"] < 0.06).mean() > 0.95,
         "complaints ≥ 0": (df["complaints"] >= 0).all(),
         "churn only Yes/No": set(df["churn"]) == {"Yes", "No"}}
for rule, ok in rules.items():
    print(("✅" if ok else "❌"), rule)

import numpy as np
import pandas as pd

def load_churn(path="telecom_churn.csv"):
    # "None" means no internet plan – stop pandas from turning it into NaN
    df = pd.read_csv(path, keep_default_na=False, na_values=[""])
    df["total_charges"] = df["total_charges"].fillna(df["monthly_charges"] * df["tenure_months"])
    df["churned"] = (df["churn"] == "Yes").astype(int)
    return df

df = load_churn()
churners = df[df["churned"] == 1]
rate = df["churned"].mean()
monthly_lost = churners["monthly_charges"].sum()
print(f"Customers: {len(df):,} | churned: {len(churners):,} | churn rate: {rate:.1%}")
print(f"Monthly revenue walking out: ₹{monthly_lost:,.0f}  → per year ≈ ₹{monthly_lost * 12 / 1e5:.1f} lakh")
print(f"Average bill – churners ₹{churners['monthly_charges'].mean():.0f} vs stayers ₹{df.loc[df['churned'] == 0, 'monthly_charges'].mean():.0f}")

import numpy as np
import pandas as pd

def load_churn(path="telecom_churn.csv"):
    # "None" means no internet plan – stop pandas from turning it into NaN
    df = pd.read_csv(path, keep_default_na=False, na_values=[""])
    df["total_charges"] = df["total_charges"].fillna(df["monthly_charges"] * df["tenure_months"])
    df["churned"] = (df["churn"] == "Yes").astype(int)
    return df

import matplotlib.pyplot as plt
df = load_churn()
fig, ax = plt.subplots(1, 2, figsize=(11, 3.8))
c = df["churn"].value_counts()
ax[0].pie(c, labels=[f"{k} ({v:,})" for k, v in c.items()], colors=["#94a3b8", "#dc2626"], autopct="%1.1f%%", startangle=90, wedgeprops=dict(width=0.45))
ax[0].set_title("29% of customers churned")
rev = df.groupby("churn")["monthly_charges"].sum() / 1e5
ax[1].bar(rev.index, rev.values, color=["#94a3b8", "#dc2626"]); ax[1].set_ylabel("₹ lakh per month")
ax[1].set_title("Monthly revenue: kept vs walked out")
for i, v in enumerate(rev.values): ax[1].text(i, v + 0.2, f"₹{v:.1f} L", ha="center")
plt.tight_layout(); plt.show()

assumptions = {
    "retention offer": "₹100 off per month for 6 months",
    "offer cost per accepted customer (₹)": 600,
    "call cost per customer contacted (₹)": 40,
    "share of true churners saved by a call + offer": 0.30,
    "share of non-churners who still take the offer": 0.50,
    "value of a saved customer (₹)": "monthly bill × 12 (one more year)",
    "call capacity per month": 600,
}
for k, v in assumptions.items():
    print(f"{k:48s} {v}")
saved_value = 759 * 12            # average churner bill ≈ ₹759 (Step 4)
print(f"\nOne saved churner ≈ ₹{saved_value:,} of revenue; cost of trying = ₹40 call (+ ₹600 if they accept)")

# TeleOne Churn Prediction – Project Charter
**Goal:** Reduce customer churn by targeting retention calls at customers most likely to leave.
**Users:** Retention team (600 calls / month).
**Deliverables:** (1) monthly risk list with top reasons, (2) Streamlit app for agents,
(3) model card and one-page report.
**Success metrics:** ROC-AUC ≥ 0.80 on unseen customers; ≥ 2× more churners reached
per call than random calling; positive net ₹ impact under agreed assumptions.
**Data:** 3,000 customers, 13 columns, snapshot at month start. No leakage columns.
**Risks:** small dataset; offer acceptance rate is an assumption → propose an A/B pilot.
**Timeline:** Day 56 problem & data · 57 EDA · 58 models · 59 evaluation · 60 deployment.