A practical comparison of four approaches for modeling insurance claim costs – Linear Regression, Two‑Part (Logistic+Gamma), Tweedie GLM, and XGBoost with Tweedie loss. Using simulated car insurance data, this project demonstrates the actuarial edge of the Tweedie distribution for zero‑inflated, skewed claim data. All models perform similarly, with XGBoost achieving a marginal 2.21% RMSE improvement. The real value lies in feature importance – revealing past claims and mileage as the strongest predictors.

A hands-on data science project comparing Linear Regression and XGBoost for retail sales forecasting. This project reveals an unexpected outcome – the simpler model wins – and explores why model selection matters more than algorithm complexity in real-world applications.

A guided walkthrough of modeling count data with excess zeros. This project compares Poisson, Negative Binomial, Zero‑Inflated (ZIP/ZINB), Hurdle, and Bayesian (JAGS) models on simulated cargo shipment data and the real bioChemists dataset. Model selection is driven by AIC and DIC, demonstrating when structural zero processes are justified – and when simpler models win. Includes practical notes on NaN AIC/DIC issues and numerical stability.