Hotdog Predictor
Predicting hotdog vendors through NLP and classification
Can a business name reveal what’s on the menu? This project explores text-based classification using the City of Vancouver’s Street Food Vending dataset.
I engineered a binary target from vendor descriptions and trained multiple models, including Decision Trees and Logistic Regression, to identify hot dog vendors based solely on their business names. After rigorous cross-validation, a tuned Naïve Bayes classifier emerged as the top performer with 79% accuracy. The project highlights the challenges of imbalanced datasets and the nuances of classifying minority target classes in urban data.
Credit
Collaborator: Built with Zaki Aslam, Hector Palafox Prieto, and Samrawit Mezgebo Tsegay