Crime & Housing Vacancy Analysis Across U.S. States

This project explored how non-violent crime rates relate to residential vacancy trends across the United States. Using datasets from the U.S. Census, FBI UCR, and the LEMAS survey, I built predictive models to identify key factors influencing housing vacancies.

The analysis involved thorough data preprocessing, including KNN imputation and custom feature engineering. I applied machine learning models—Random Forest Classification, Gradient Boosting, and Linear Regression—to evaluate the impact of variables like homelessness and drug-related arrests. Homelessness was found to significantly improve model performance (+2.9%).

Key results included:
• Random Forest: 69.3% accuracy, 69.4% F1-Score, 84.7% ROC-AUC
• Linear Regression: Explained 48.4% of the variance in vacancy rates

The project emphasized strong collaboration, statistical insight, and iterative model tuning.

Tools & Skills: Python, Scikit-learn, Pandas, Matplotlib, ETL, EDA, Predictive Modeling, Feature Engineering