https://github.com/jgavinb/banking-insurance-purchase-prediction-mars-gam
https://github.com/jgavinb/banking-insurance-purchase-prediction-mars-gam
Last synced: 3 months ago
JSON representation
- Host: GitHub
- URL: https://github.com/jgavinb/banking-insurance-purchase-prediction-mars-gam
- Owner: jgavinb
- Created: 2025-02-20T20:19:34.000Z (3 months ago)
- Default Branch: main
- Last Pushed: 2025-02-20T20:20:53.000Z (3 months ago)
- Last Synced: 2025-02-20T21:19:33.737Z (3 months ago)
- Size: 0 Bytes
- Stars: 0
- Watchers: 1
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
Awesome Lists containing this project
README
# Insurance Purchase Prediction with MARS and GAM
## Description
This project predicts customer purchases of a variable annuity insurance product using **MARS** (Multivariate Adaptive Regression Splines) and **GAMs** (Generalized Additive Models). The solution demonstrates expertise in handling missing data, advanced regression techniques, and model evaluation for a commercial banking use case.
**Tools Used**
- **R**: `earth`, `mgcv`, `caret`, `pROC`, `dplyr`
- **Visualization**: `ggplot2`, `ROCit`
- **Workflow**: RMarkdown for reproducible reporting
## Key Features
- Imputed missing values using **median** for continuous variables and **new category (2)** for missing binary/categorical data.
- Built a **MARS model** with binomial logistic regression (`glm.list(family="binomial")`).
- Developed a **GAM** with splines for continuous variables and automated feature selection (`select = TRUE`).
- Evaluated models using **AUC-ROC curves** and interpreted variable importance.
## Results
- MARS achieved **AUC-ROC of 0.85**, identifying **account age (ACCTAGE)** as a key predictor.
- GAM revealed nonlinear relationships with **home value (HMVAL)** and **credit score (CRSCORE)** driving purchase decisions.
- Delivered actionable insights through visualizations and business-ready documentation.