Tackling Noise in Ranking Models with the Boosting Paradigm

Raanking data are used in many fields to analyze preferences, but noise in the data---such as random or inconsistent rankings---can make interpretation tricky. This paper tackles the issue by introducing a boosting-based approach to improve the Mallows-Bradley-Terry model, making it more resistant to noise. The method works by iteratively adjusting weights based on how well individual rankings align with an estimated consensus. Through Monte Carlo simulations, we show that this approach leads to clearer, more consistent probabilities, even when the data contains a high percentage of noise.