Vinka Predicts Transportation Demand with Machine Learning on Google Cloud

Vinka Predicts Transportation Demand with Machine Learning on Google Cloud


What was the challenge?

Vinka  is a software and SaaS platform company specializing in personalized transportation solutions for mobility service providers. Vinka Ride is a booking and dispatch platform for automating and optimizing taxi and on-demand fleets. It is easy to use and has powerful optimization features including ETA based dispatch optimization and automated ride-pooling. Managing priorities and different types of requirements within the same operation is a complex challenge. 

Vinka has collected order data for several years. The data shows, among other things, the coordinates of the starting point of the taxi order and the desired departure time of the taxi. However, this data is not yet used for predicting transportation order volumes. Vinka wanted to see if the existing order data could be used for more accurate forecasts of future demand in certain locations dynamically.

The most significant challenge was predicting order quantities simultaneously in time and place. Achieving these two goals simultaneously creates problems for a large part of data analytics tools. A demand forecast calculated for an individual area must be accurate to be useful, but at the same time the preparation of the forecast must be light enough so that these forecasts can be efficiently prepared for a large number of areas.


Our solution

Codento developed a solution that utilizes artificial intelligence and location data, which can be used to predict the number of orders simultaneously in time and place. The solution used Google Cloud’s BigQuery tool. BigQuery natively supports both machine learning models (BigQuery ML and Vertex AI AutoML) and geospatial filtering (BigQuery GIS). By combining these two tools, Codento managed to build an efficiently scalable forecasting model that can simultaneously forecast the demand for taxis for a practically unlimited number of different geographical areas. Moreover, the shapes of all these areas can be defined completely freely.


Why Codento?

Codento was chosen for the collaboration based on extensive and versatile artificial intelligence consulting experience and Google’s cloud expertise. Codento’s earlier similar implementations also had a positive effect on the decision.


What were the results?

With the help of BigQuery ML, it was possible to build a model that is able to predict the number of orders coming in an hour in a pre-defined geographical area up to 10 days in advance with amazing reliability. In addition, the model was able to take into account the effect of the public holidays on demand quite reliably. The biggest positive outcome, however, was the scalability of the model. The machine learning model was able to be trained for a single geographical area in about ten seconds. This short training time allows the model to be effectively scalable


In customer’s own words

Codento was able to prove that a combination of personal transportation order data and geospatial location data can be combined into an accurate demand forecast using a machine learning model powered by Google Cloud BigQuery.” – Peitsa Turvanen, CEO, Vinka

by Peitsa Turvanen, CEO, Vinka


Please contact  us for more information about our services: