I am not sure how much of an edge one can gain by doing demand prediction through algorithms. In large markets the patterns are fairly trivial (big game ended, everybody goes home; it's 5 pm in financial district, people are leaving work; busy time at the airport, bunch of flights arriving at once) and known to all participants.
My point was not about (officially or not) scheduled spikes in traffic. These are known for all parties in advance. I was talking more about organic traffic - Random Joe hailing a cab just because. And cabs generally just drive around expecting to catch the Random Joe wasting time/vehicle/fuel. Now if you can predict temporal/geographical density of Random Joes hailing a cab - well, you have an edge there simply by increasing utilisation rate. Is this even possible? Have no idea.
