Telco Customer Churn Prediction

📌 Project Objective

The goal of this project is to predict the likelihood of a Telco customer churning using Machine Learning techniques with Python and TensorFlow. This allows the company to implement retention strategies and reduce customer loss.

---

🗂 Project Structure

Telco_Churn/
├─ data/
│ ├─ raw/ # Original dataset downloaded from Kaggle
│ ├─ processed/ # Cleaned and transformed data
├─ notebooks/
│ └─ EDA.ipynb # Initial exploratory data analysis
├─ src/
│ ├─ preprocessing.py # Data cleaning and preprocessing functions
│ ├─ model.py # Model definition and training
│ ├─ evaluate.py # Model evaluation and metrics
│ └─ visualize.py # Visualization functions (Precision-Recall curve)
├─ outputs/
│ ├─ figures/ # Generated plots
│ └─ models/ # Saved trained models (.keras)
├─ requirements.txt # Project dependencies
├─ README.md # Project documentation
└─ run.py # Main script to execute the full pipeline

---

🛠 Tools and Libraries

• Python 3.x
• pandas, numpy
• scikit-learn
• TensorFlow
• matplotlib

---

📈 Pipeline Overview

1. Data Loading & Preprocessing

   • Load raw dataset and clean missing values.
   • Convert categorical variables using one-hot encoding.
   • Scale numerical features with StandardScaler.

2. Model Training

   • Neural network built with TensorFlow/Keras.
   • Dense layers with ReLU activation and Dropout for regularization.
   • Binary classification using sigmoid output.

3. Evaluation

   • Compute Precision, Recall, F1-score and ROC-AUC.
   • Calculate the best threshold to maximize F1-score.

4. Visualization

   • Precision-Recall curve is saved in outputs/figures/.
   • Threshold that maximizes F1 is highlighted on the plot.

5. Outputs

   • Trained model saved in outputs/models/.
   • Predictions can be exported for further analysis or Power BI dashboards.

---

📊 Metrics Example

After training, example results:

• Accuracy: 0.76
• ROC-AUC: 0.82
• Best F1 Threshold: 0.54
• Class 1 (Churners): Precision 0.53, Recall 0.74, F1 0.62

Note: Metrics are computed on a hold-out test set. The threshold is optimized for business use-case to capture most churners while controlling false positives.

---

🚀 How to Run

1. Clone the repository:

git clone https://github.com/quantumRizo/Telco_Churn.git