🧠 Customer Churn Prediction

This project builds an end-to-end deep learning pipeline to predict customer churn using an Artificial Neural Network (ANN). It includes preprocessing, feature engineering, hyperparameter tuning, model evaluation with TensorBoard, and deployment using Streamlit.

---

🚀 Problem Statement

Customer churn is a major problem in industries like telecom, where retaining existing customers is more cost-effective than acquiring new ones. This project aims to:

• Identify patterns leading to customer churn.
• Predict churn using historical data.
• Provide actionable insights for business decision-making.

---

🧭 How It Works

🔄 Project Workflow

1️⃣ Data Exploration & Visualization (EDA)

• 📊 Univariate & Bivariate Analysis
  → Explored features like gender, tenure, and contract type
• 🔥 Churn Distribution & Correlation Analysis
  → Used heatmaps to identify feature relationships and potential churn indicators

---

2️⃣ Preprocessing

• 🧩 Categorical Encoding
  → Applied Label Encoding & One-Hot Encoding where appropriate
• 📏 Feature Scaling
  → Used StandardScaler to normalize numerical inputs
• 💾 Saving Pipelines
  → Persisted encoders & scalers using pickle for reuse in app deployment

---

3️⃣ Modeling

• 🔍 Hyperparameter Tuning
  → Optimized with GridSearchCV to find best batch size, epochs, and optimizer
• 🧠 Model Architecture
  → Built an Artificial Neural Network (ANN) using TensorFlow/Keras
• 📉 Evaluation & Monitoring
  → Integrated EarlyStopping and TensorBoard for training visualization and diagnostics

4️⃣ 📈 Model Training Metrics (TensorBoard)

Below are the training metrics visualized using TensorBoard during model training:

📊 Key Observations

• ✅ Validation Accuracy steadily improved, reaching ~86.9%
• 📉 Validation Loss consistently decreased, indicating effective learning
• 🧠 Training vs. Validation curves show no overfitting, thanks to EarlyStopping

---

5️⃣ 🌐 Deployment

• 🚀 Deployed via a Streamlit app for real-time, interactive churn prediction
• 🧩 Integrated with saved preprocessing pipeline and trained ANN model

---

🛠️ Tech Stack

• Languages & Libraries: Python, Pandas, NumPy, Matplotlib, Seaborn
• Machine Learning: Scikit-learn, TensorFlow, Keras
• Hyperparameter Tuning: GridSearchCV
• Visualization & Monitoring: TensorBoard
• Deployment: Streamlit
• Platform: Google Colab + GitHub

---