Here, you can see a sharp difference between the morning hourly wage for good and bad drivers. The algorithm learns from the behaviour of good drivers.
Our algorithm determines where a taxi driver should search for customers by looking at the behaviour of past good drivers, factoring in real driving time data and the cost of gas. Poorly preforming drivers realize about a 10% increase in their average morning hourly wage if they follow our machine-learning based strategy, as you can see here.
Here, you can see clearly why drivers that follow our more informed strategy make more money. Drivers that follow the naive strategy tend to make short trips to locations nearby. In comparison, drivers that follow the infomed strategy often have the audacity to make one or two trips without a customer in hopes of finding a more profitable customer (represented below with a blue line). Typical trajectories are plotted below (here, the driver starts at the blue rhombus).
Here, you can study heatmaps of locations frequented by good and bad drivers, or download all of our code from GitHub to train our machine learning algorithm and run these simulations yourself. They are available via the links below.Good Drivers Poorly Preforming Drivers