DEEP-CSAT — eCommerce Customer Satisfaction Prediction

Project Type: Binary Classification
Target: Predict whether a customer is Satisfied (CSAT 4–5) or Not Satisfied (CSAT 1–3)
Dataset: eCommerce_Customer_support_data.csv (~83K records, 20 columns)

---

📁 Project Structure

DEEP-CSAT Project/
├── eCommerce_Customer_support_data.csv  ← raw dataset
├── deep_csat_pipeline.py                ← full ML pipeline (train + save model)
├── app.py                               ← Flask REST API (serve predictions)
├── requirements.txt                     ← Python dependencies
├── deep_csat_model.pkl                  ← saved model (generated after training)
├── plots/                               ← 15+ EDA & model charts (auto-generated)
└── README.md

---

⚙️ Setup

# 1. Create & activate virtual environment
python -m venv venv
venv\Scripts\activate          # Windows
# source venv/bin/activate     # macOS / Linux

# 2. Install dependencies
pip install -r requirements.txt

---

🚀 Run the Full ML Pipeline

python deep_csat_pipeline.py

This will:

1. Load & explore the dataset
2. Engineer features (response time, temporal, NLP/TF-IDF)
3. Generate 15 EDA charts → saved to plots/
4. Run 3 hypothesis tests (t-test, chi-square, ANOVA)
5. Handle class imbalance (SMOTE)
6. Train 3 ML models (Logistic Regression, Random Forest, XGBoost)
7. Save best model → deep_csat_model.pkl
8. Run a sanity check on an unseen sample

---

🌐 Run Flask API (Local)

Train the pipeline first to generate deep_csat_model.pkl

python app.py

Predict endpoint

POST http://localhost:5000/predict

{
  "channel_name": "Inbound",
  "category": "Product Queries",
  "Sub-category": "Product Specific Information",
  "Customer_City": "Mumbai",
  "Product_category": "Electronics",
  "Agent Shift": "Morning",
  "Tenure Bucket": ">90",
  "Item_price": 1500.0,
  "connected_handling_time": 8.0,
  "response_time_min": 12.0,
  "issue_hour": 10,
  "issue_dayofweek": 1,
  "issue_month": 8,
  "Customer Remarks": "product working great happy service"
}

Response:

{
  "prediction": 1,
  "label": "Satisfied",
  "confidence": 0.8743,
  "confidence_pct": "87.43%"
}

---

☁️ Cloud Deployment (Render / Railway / Heroku)

1. Push project to GitHub
2. Add a Procfile:

   web: gunicorn app:app
   

3. Set Start Command to python deep_csat_pipeline.py && gunicorn app:app
   (trains on first deploy, then serves)
4. Set Python version to 3.10

---

📊 Model Performance Summary

Model	Accuracy	ROC-AUC
Logistic Regression	~0.72	~0.77
Random Forest	~0.79	~0.84
XGBoost ✅	~0.81	~0.87

XGBoost selected as final model for deployment.

---

🔑 Key Features

Feature	Importance
response_time_min	High
connected_handling_time	High
Tenure Bucket	Medium
Agent Shift	Medium
channel_name	Medium
NLP TF-IDF terms	Variable

---

📋 Notebook Submission

Open Sample_ML_Submission_Template-2.ipynb in Google Colab or Jupyter.
The pipeline code in deep_csat_pipeline.py maps 1-to-1 to the notebook sections.