UnFare
A flight-price monitor that flags unusually low fares using statistical models, route clustering, and airline-specific pricing baselines.
I built UnFare to look for unusually low flight prices without treating every cheap ticket as a mistake fare. The detector combines statistical outlier scores, route-level clustering, and rules based on historical and airline-specific pricing.
A feature model considers price, route distance, seasonality, stops, and cabin class. Clustering compares a fare with similar routes, and a rule layer combines the results into a candidate alert. Feature perturbation helps me inspect which inputs drive a score and investigate false positives.
Playwright collects prices from airline and travel-booking sites, with proxy rotation and route-level circuit breakers for blocked endpoints. TimescaleDB stores price history alongside fare classes and booking URLs. The output identifies fares worth checking; a low price alone does not establish an airline error.