Customer churn prediction is a critical task in business analytics aimed at identifying customers likely to discontinue a service or product. This project leverages machine learning models to predict customer churn using the Churn Modelling.csv dataset, encompassing demographic, financial, and behavioral features. The project is divided into six key parts:

1. Data Cleansing: Preprocessing, missing value handling, encoding categorical variables, and handling outliers.
2. KNN Model: Using the k-Nearest Neighbors algorithm for churn prediction.
3. Decision Tree, Random Forest, and Gradient Boosting Models: Comparing the performance of these models.
4. Support Vector Machine (SVM): Using different kernels to improve prediction.
5. Neural Networks: Implementing artificial neural networks for classification.
6. Model Comparison: Evaluating and comparing all models for the best performance.

• Handled missing values and irrelevant columns.
• Encoded categorical variables using one-hot encoding.
• Addressed class imbalance using SMOTE (Synthetic Minority Oversampling Technique).

• k-Nearest Neighbors (KNN):
  
  • Achieved optimal results with k = 9 using cross-validation.
• Decision Tree, Random Forest, and Gradient Boosting:
  
  • Hyperparameter tuning with GridSearchCV for maximum efficiency.
• Support Vector Machines (SVM):
  
  • Explored linear, RBF, polynomial, and sigmoid kernels.
• Neural Networks:
  
  • Compared solvers for weight optimization and activation functions.

• Conducted a detailed comparison using:
  
  • Accuracy
  • Precision
  • Recall
  • F1-Score
  • ROC Curve and AUC values

• Programming Languages: Python
• Libraries:
  
  • Preprocessing: pandas, numpy
  • Visualization: matplotlib, seaborn
  • Machine Learning: scikit-learn, imbalanced-learn
• Techniques: Hyperparameter tuning, SMOTE, cross-validation, feature importance analysis

• Best Performing Model: Random Forest with an AUC of 0.86 and strong overall predictive accuracy.
• Feature Importance:
  
  • Age and NumOfProducts were consistently the most important predictors across models.
• Insights:
  
  • Addressed challenges like class imbalance and overfitting with techniques such as SMOTE and pruning.

├── Data/ │ ├── Churn_Modelling.csv ├── Analysis/ │ ├── scripts/ │ │ ├── data_cleansing.py │ │ ├── knn_model.py │ │ ├── decision_tree_model.py │ │ ├── svm_model.py │ │ ├── neural_network_model.py ├── Visualizations/ │ ├── roc_curves/ ├── Reports/ │ ├── Customer_Churn_Analysis.pdf ├── README.md

For a comprehensive understanding of the project, including detailed methodologies, visualizations, and results, refer to the full report: Customer Churn Analysis Report.

If you’d like to contribute, suggest improvements, or have any questions, feel free to open an issue and please reach out via LinkedIn

Author: Syed Faizan
Master’s Student in Data Analytics and Machine Learning