This repository contains the work completed for the job simulation involving customer churn prediction for the Data Science & Analytics team at Lloyds Banking Group. The objective of the project was to develop a predictive model that identifies customers at risk of churning, enabling the business to implement effective retention strategies.
- Developed and implemented a predictive model using Random Forest and other machine learning algorithms, achieving an F1 score of 0.862.
- Conducted advanced data preprocessing, including:
- Handling missing values
- Encoding categorical variables
- Feature scaling
- Utilised Python libraries such as pandas, scikit-learn, and matplotlib for data manipulation, model building, and visualisation.
- Performed comprehensive model evaluation and tuning by:
- Optimising hyperparameters with GridSearchCV
- Analysing feature importance to derive actionable business insights.
The final model achieved an F1 score of 0.862, indicating a strong ability to balance precision and recall in predicting customer churn. Key features impacting churn prediction were identified, providing actionable insights for customer retention strategies.
Explore additional modeling techniques (e.g., ensemble methods or neural networks) for potentially improved performance. Continuously update the model with new data to enhance accuracy and adapt to changing customer behaviors. Develop an interactive dashboard for real-time monitoring and decision-making based on churn predictions.
Thank you to Lloyds Banking Group and Forage for the opportunity to participate in this job simulation and enhance my skills in data science and analytics.
Data Inspection
- Variable Analysis Exploratory Data Analysis Datasets:
- Customer Demogrpahics
- Transactional History
- Customer Service
- Online Activity Data Cleaning
- Check Null Values
- Handle Missing Values
- Merge Datasets
- Further EDA on new columns
- Statiscal Analysis testing
Final Inspection on cleaned data
- Statistical Analysis on features to determine standardisation, normalisation methods for feature engineering
- Q-Q test to analyse normality of features Feature Engineering
- Encoding using Onehot, Ordinal Encoder
- Scaling using StandardScaler
- Transformation (log)
- Preproccessing & building pipelines Modelling
- Split into Train & Test
- SMOTE on imbalanced data
- Fit & Compare Models (logistic regression, decision trees, random forest, gradient boosting, neural network Model Metrics (accuracy, precision, ROC-AUC, recall, f1-score Hyperparameter Testing (f1 score of random forest increased from 0.24 to 0.862 using GridSearchCV)
