Churn Prediction with Machine Learning

This repository contains a machine learning project focused on predicting customer churn using a dataset that includes various customer attributes and service usage information. The project involves data preprocessing, exploratory data analysis, feature engineering, model selection, and evaluation.

📌 Project Objectives

• Predict whether a customer is likely to churn or not.
• Perform exploratory data analysis (EDA) to understand key churn indicators.
• Apply feature engineering to enhance model performance.
• Train and compare multiple machine learning models.
• Evaluate model performance using appropriate metrics.

📁 Dataset

The dataset contains 7,043 rows and 21 variables related to customer demographics, services signed up for, and account information.
It is a simulated Telco dataset, widely used in churn prediction exercises.

• Categorical, numerical, and binary variables included
• Target variable: Churn

🛠️ Technologies & Libraries

• Python
• pandas, numpy — for data manipulation
• matplotlib, seaborn — for data visualization
• scikit-learn — for machine learning models and evaluation
• XGBoost, LightGBM, CatBoost — advanced gradient boosting frameworks

🔍 Project Workflow

1. Data Preprocessing
   • Handling missing values
   • Encoding categorical variables
   • Outlier detection and treatment (if needed)
2. Exploratory Data Analysis (EDA)
   • Analyzing churn distribution
   • Investigating relationships between features and churn
3. Feature Engineering
   • Creating new meaningful variables
   • Encoding binary/multiclass features
4. Modeling
   • Logistic Regression
   • K-Nearest Neighbors
   • Decision Tree
   • Random Forest
   • XGBoost
   • LightGBM
   • CatBoost
   • Hyperparameter tuning with GridSearchCV
5. Model Evaluation
   • Accuracy, Precision, Recall, F1-score
   • ROC-AUC
   • Cross-validation
   • Model comparison

📊 Results

Model	Accuracy	F1-Score	ROC-AUC
Random Forest	0.7994	0.5648	0.8454
XGBoost	0.8007	0.5822	0.8448
LightGBM	0.8031	0.5890	0.8454
CatBoost	0.8048	0.5860	0.8469

• Among 8 tested algorithms, CatBoost achieved the best performance with a ROC-AUC score of 0.8469 and an F1-score of 0.5860, demonstrating strong predictive power for customer churn.