I was recently interested in road traffic speed prediction. The problem with academic research into predicting speeds is not as much the lack of reproducibilit

I was recently interested in road traffic speed prediction. The problem with academic research into predicting speeds is not as much the lack of reproducibility, but the lack of measures that make sense. Hundreds of papers compare a short-term prediction (say 30 minutes ahead) using mean average error, or similar measures. This completely ignores the fact that vehicle speeds are mostly constant and only vary significantly during rush hours or other disruptions, at which time the process of breakdown is rather chaotic. From what I gather this will not change easily, because it is considerable academic fun to improve on a number for existing benchmarks, instead of redefining the game. I'd love to help academics in this field who need some guidance on practical applications!