---

🧠 About This Project

project = {
    "name"        : "BCG X Data Science Job Simulation",
    "platform"    : "Forage",
    "company"     : "BCG X — Boston Consulting Group's Tech & Digital Ventures",
    "completed"   : "March 23rd, 2026",
    "client"      : "PowerCo — SME Gas & Electricity Provider",
    "tasks"       : 5,
    "dataset"     : "14,606 SME Customers | Energy Consumption & Pricing Data",
    "tools"       : ["Python", "Pandas", "NumPy", "Scikit-Learn", "Matplotlib", "Seaborn"],
    "skills"      : ["EDA", "Feature Engineering", "Random Forest", "ROC-AUC",
                     "Business Hypothesis Testing", "SCQA Framework", "Executive Reporting"],
    "outcome"     : "🏆 Certificate Issued — Data Science Job Simulation"
}

💡 "PowerCo thought price was killing their business. I let the data speak — and it told a completely different story."

---

📊 Key Metrics

👥 Total Customers	🚨 Churned	📉 Churn Rate	🤖 Model ROC-AUC
14,606	1,419	9.7%	0.71

---

📁 Repository Structure

📦 BCG-Data-Science-Job-Simulation/
│
├── 📓 BCG_Task3_EDA.ipynb                    → Exploratory Data Analysis
├── 📓 BCG_Task4_Feature_Engineering.ipynb    → Feature Engineering
├── 📓 BCG_Task5_Modeling.ipynb               → Random Forest Modeling
├── 📄 BCG_Executive_Summary.pdf              → SCQA Business Report
├── 📊 churn_distribution.png
├── 📊 consumption_analysis.png
├── 📊 correlation_heatmap.png
├── 📊 feature_importance.png
├── 📊 margin_analysis.png
├── 📊 model_evaluation.png
├── 📊 price_analysis.png
├── 📊 tenure_analysis.png
└── 📄 README.md

⚠️ Note: Raw data files excluded per BCG X confidentiality policy.

---

🔬 Executive Summary — SCQA Framework

📍 SITUATION
└── PowerCo has a 9.7% churn rate (1,419 of 14,606 customers)
    Churned customers have HIGHER avg margin (€228 vs €185)
    → The business is losing its most valuable clients first

⚠️ COMPLICATION
└── PowerCo hypothesised price sensitivity as the primary churn driver
    Analysis shows price is NOT the key driver
    → Consumption, margin & tenure are stronger predictors

❓ QUESTION
└── Is price sensitivity the primary driver of SME churn at PowerCo?
    Can a targeted discount strategy reduce churn while protecting margins?

✅ ANSWER
└── Random Forest model (ROC-AUC: 0.71) identifies at-risk customers
    Offer 20% discounts ONLY to high-margin, high-consumption customers
    Focus on early-tenure customers (1–3 yrs) — ~27% churn rate
    Refine model recall before full rollout

---

🎯 Task 3 — Exploratory Data Analysis

Methodology

Step	Action
🧹 Data Cleaning	Handled missing values, outliers, data type conversions
📊 Churn Analysis	Distribution of churned vs retained across all features
⚡ Consumption	Electricity & gas usage patterns by churn status
💰 Price Analysis	Off-peak variable & fixed price distributions
📈 Margin Analysis	Net margin comparison — churned vs retained
📅 Tenure Analysis	Churn rate by years with PowerCo
🔥 Correlation	Feature correlation heatmap

🔍 Key EDA Findings

📌 9.7% churn rate          → 1,419 of 14,606 customers churned
📌 Churned avg margin €228  → vs €185 retained — losing best clients!
📌 New customers (1-2 yrs)  → ~27% churn rate — 3x the average
📌 Price distributions      → nearly identical for churned vs retained
📌 Consumption patterns     → clearly differ between churned & retained
📌 cons_12m & cons_last_month → 0.97 correlation — highly related

📸 EDA Visualizations

---

⚙️ Task 4 — Feature Engineering

Methodology

Step 1 → Price variability features (off-peak vs peak mean differences)
Step 2 → Tenure-based features (months active, months since product change)
Step 3 → Consumption ratio features (last month vs 12-month avg)
Step 4 → Margin-based features (gross vs net power electricity margin)
Step 5 → Final feature selection → exported as final_features.csv

Key Engineered Features

Feature	Description
off_peak_peak_var_mean_diff	Price variability between off-peak & peak periods
off_peak_mid_peak_var_mean_diff	Price variability between off-peak & mid-peak
months_activ	Number of months customer has been active
months_modif_prod	Months since last product modification
var_year_price_off_peak	Year-on-year off-peak price change

---

🤖 Task 5 — Modeling & Evaluation

Methodology

Step 1 → Train/Test Split (80/20)
Step 2 → Handle class imbalance
Step 3 → Random Forest Classifier training
Step 4 → ROC-AUC evaluation
Step 5 → Feature importance extraction
Step 6 → Business interpretation of results

📊 Model Results

Metric	Score
ROC-AUC	0.706
True Negatives (Correctly Retained)	2,635
False Negatives (Missed Churners)	260

🏆 Top 15 Feature Importances

Rank	Feature	Importance Score
🥇 1	cons_12m — 12-month electricity consumption	0.0525
🥈 2	margin_net_pow_ele — Net power electricity margin	0.0524
🥉 3	margin_gross_pow_ele — Gross power electricity margin	0.0519
4	forecast_meter_rent_12m — Forecasted meter rent	0.0502
5	net_margin — Overall net margin	0.0448
6	forecast_cons_12m — Forecasted consumption	0.0440
7	cons_last_month — Last month consumption	0.0372
8	pow_max — Max power subscribed	0.0333
9	months_activ — Months active	0.0330
10	months_modif_prod — Months since product change	0.0312

🔑 Price features ranked well below consumption & margin — confirming price is NOT the primary churn driver!

📸 Model Visualizations

---

💡 Business Recommendations

#	Recommendation	Data Behind It
1️⃣	Do NOT apply blanket 20% discounts	Price is NOT the primary churn driver
2️⃣	Target high-margin + high-consumption customers	Top features in Random Forest model
3️⃣	Focus on 1–3 year tenure customers	~27% churn rate — 3x the average
4️⃣	Improve model recall before full rollout	Current model misses some churners
5️⃣	Use RF model to proactively flag at-risk customers	ROC-AUC: 0.71 — good discriminatory power

---

🛠️ Tech Stack

---

💼 Skills Demonstrated

Analytics	Machine Learning	Business
✅ End-to-end EDA	✅ Random Forest Classifier	✅ SCQA Framework Reporting
✅ Feature Engineering	✅ ROC-AUC Evaluation	✅ Executive Summary Writing
✅ Data Visualisation	✅ Feature Importance Analysis	✅ Hypothesis Testing
✅ Correlation Analysis	✅ Class Imbalance Handling	✅ Business Recommendations

---

🏆 Certificate of Completion

Field	Details
🏅 Certificate	Data Science Job Simulation
🏢 Issued By	BCG X via Forage
📅 Completed	March 23rd, 2026
🔗 View Certificate	Click Here

---

👨‍💻 Author

Devesh Shukla Data Analyst | ML Enthusiast | Insight Storyteller

⭐ If you find this useful, please give it a star! ⭐