Arthur Motta
About Me
I am an undergraduate student in Actuarial Science and Statistics at UFRJ (Universidade Federal do Rio de Janeiro), focused on actuarial modeling and data science applied to the insurance industry.
My work combines classical actuarial methods - GLM frequency-severity models, credibility theory, loss ratio analysis - with modern machine learning techniques such as XGBoost, SHAP explainability, and predictive modeling on large-scale regulatory datasets.
Projects
-
Python · GLM Poisson · GLM Gamma · XGBoost Tweedie · SHAP · SUSEP AUTOSEG 2019-2021End-to-end actuarial pricing model built on 12.6 million real policy records from Brazil's insurance regulator (SUSEP). Models collision and theft as separate risks — their claim frequency correlation is only 0.021, confirming distinct risk drivers. Uses GLM Poisson × Gamma for the standard actuarial freq-sev decomposition (Gini 0.241 for collision, 0.402 for theft), with XGBoost Tweedie as a benchmark for the zero-inflated pure premium distribution. SHAP explainability was incorporated to ensure model transparency and regulatory interpretability — a requirement for SUSEP filings. Market premium analysis compares the modeled pure premium against SUSEP's reported commercial premium, revealing relative over and underpricing across 41 regions and five policyholder profiles.