Predicting driver density in Bogotá
Whilst working at a start-up in Colombia I was tasked with building a machine learning system to predict where drivers are needed across the city at any given time.
This involved re-engineering historic data to aggregate densities of drivers spatiotemporally (a grid cell location system of dynamic granularity was created using a method based on geohashing). The XGBoost algorithm was used to build the model, and an API built to deploy it. One main function was to predict densities across the whole city for a heatmap display, built with Dart and used to visualise real-time and future predictions. Specific location suggestions for individual drivers could also be requested, using a matching algorithm developed to consider the driver's distance to high density locations along with the locations of other drivers in the city.