VahanBima CLTV Prediction | ML Hackathon

Problem Definition

The goal of this project is to build a robust machine learning pipeline to predict Customer Lifetime Value (CLTV) for an insurance provider. Accurate CLTV prediction enables data-driven decisions regarding customer retention strategies, premium structures, and high-value customer acquisition.

Dataset Profile

• Scale: 89,392 records in the training set with zero missing values across all features.
• Target Feature (cltv): A highly right-skewed numerical target with a baseline skewness of 2.75. The median value is 66,396, while extreme outliers stretch up to 724,068.
• Primary Metric: Evaluation is strictly bound to the $R^2$ Score on the original scale.

Notebook Workflow

1. Initial Data Inspection and Preprocessing
2. Exploratory Data Analysis (EDA)
3. Feature Engineering
4. OOF target encoding
5. Model Development & Hyperparameter Optimization (CatBoostRegressor & Optuna)
6. Final Tuned Model Training and Prediction