Telecom Churn Predictor

Overview

This project focuses on predicting customer churn using machine learning models. By analyzing a dataset of customer behaviors and service usage, the project identifies key patterns and provides actionable insights to help businesses retain at-risk customers. The primary objective is to minimize churn by proactively targeting customers likely to leave.

---

Objectives

• Churn Prediction: Predict customers likely to churn using advanced machine learning models.
• Customer Insights: Identify key characteristics and behaviors that correlate with churn.
• Business Impact: Enable proactive retention strategies by prioritizing high-risk customers.

---

Dataset

The dataset consists of 5,000 customer records, including:

• Categorical Features: state, international_plan, voice_mail_plan.
• Numerical Features: Usage metrics like total_day_minutes, total_night_calls, and more.
• Target Variable: Churn (binary classification: 1 for churn, 0 for no churn).

The dataset is clean, with no missing values and features already preprocessed.

---

Methodology

Data Preprocessing

1. Dropped redundant features (e.g., phone_number).
2. Encoded categorical variables using LabelEncoder.
3. Balanced the target variable using class weights and stratified sampling.

Exploratory Data Analysis (EDA)

• Visualized customer distribution across features.
• Analyzed correlations to identify relationships between variables and churn.

Machine Learning Models

The following models were trained and evaluated:

1. Logistic Regression
   • Scaled features using StandardScaler.
   • Class weights handled imbalance in Churn.
2. Random Forest
   • Leveraged for feature importance and robust performance.
   • No scaling required due to tree-based nature.
3. XGBoost
   • Tuned scale_pos_weight to address class imbalance.
   • Achieved the highest recall and best overall performance.

---

Key Results

• Best Model: XGBoost achieved a recall of 0.84, outperforming other models.
• Feature Importance:
  • Key drivers of churn include international_plan, number_customer_service_calls, and voice_mail_plan.
• Churn Strategy: Ranked top 500 customers with the highest churn probability for proactive engagement.

---

Tools and Technologies

• Programming: Python (Pandas, NumPy, Scikit-learn, XGBoost).
• Visualization: Matplotlib, Seaborn.
• Model Evaluation: Confusion matrix, ROC curve, classification report.
• Cross-Validation: Recall as the primary metric for imbalanced classification.

---

Visualizations

• Correlation Matrix: Explored relationships between features.
• Feature Importance: Highlighted key predictors of churn.
• Customer Segmentation: Visualized churn probabilities by feature subsets.

---

Future Work

• Hyperparameter tuning for optimal model performance.
• Deploy the model for real-time churn prediction.
• Integrate external factors (e.g., customer demographics) for enhanced insights.

---

Installation and Usage

1. Clone the repository:

   git clone https://github.com/your-username/churn-prediction.git
   cd churn-prediction