Data Science Customer Analytics EDA

Customer Churn Prediction for SmartBank

Data understanding & EDA: Explored customer distributions by tenure, products, spend, complaints and churn flag; visualised class imbalance and key numeric relationships (e.g. tenure vs churn rate).
Cleaning & preprocessing: Handled missing values, normalised numeric features where appropriate and encoded categorical variables (e.g. product categories, geography) using one-hot encoding.
Feature engineering: Created behaviour-based features such as: total products held, complaint frequency, average monthly spend, recency of last transaction and engagement indicators (e.g. digital vs branch-heavy usage).
Class imbalance handling: Addressed the lower proportion of churners using techniques such as class weighting and resampling to avoid models simply predicting the majority class.
Modeling: Started with a Logistic Regression baseline, then experimented with tree-based models (e.g. Random Forest, Gradient Boosting) to capture nonlinear interactions and feature importance.
Evaluation: Assessed models using metrics aligned with churn use cases: precision, recall, F1 score and ROC-AUC, with particular emphasis on recall for churners to minimise false negatives.
Explainability: Used feature importance plots and partial interpretations of key drivers (e.g. high complaints + declining spend) to inform business actions.

Working on a churn prediction model for SmartBank, a fictional Lloyds Banking Group subsidiary, as part of the Data Science virtual experience on Forage. The project combines EDA, feature engineering and machine learning to flag at-risk customers for targeted retention strategies.

2025 · Lloyds - Data Science Job Simulation (Forage)

Domain: Banking · Customer Retention

Focus: Early identification of high-risk churn customers

🚧 Project in progress: This case study is actively being developed as part of the Lloyds Banking Group Forage job simulation. Data preparation, modelling, and evaluation sections will be added once analysis is complete.