TrustWheel

Used-car buyers rarely have a reliable, data-backed way to tell whether an asking price is fair, or whether a listed mileage and condition line up with what similar cars actually sell for. Most online tools rely on generic, opaque pricing guides rather than transparent models trained on real transaction data.

• Sourced and cleaned a large real-world dataset of Maryland MVA vehicle sales records, handling missing fields, inconsistent formatting, and outliers.
• Engineered features from make, model, year, mileage, and condition to feed a set of regression models predicting fair market value.
• Trained and compared multiple regression and classification approaches (including ensemble methods) to both price vehicles and flag anomalies — cars priced well below or above what the model expects.
• Built a command-line interface so a user can input a vehicle's details and get an instant valuation, mileage forecast, and a 'good deal' flag, without needing to touch the underlying model code.

The result is a working analytics CLI that turns a public government dataset into a practical tool for evaluating used-car listings — combining real data engineering with applied machine learning end to end.