Predicting riverbed depth using Linear Regression
Predicting the shortest depth of a river is of crucial importance. With the least surveyed depth (LSDs), shipping determines how much cargo they can transport along the river. By using Machine Learning on historical data, we were able to predict future river depth. We made very accurate predictions for 2, 4 and 6 weeks ahead.
This project presented several data challenges:
- The data size approximately 500 GB, which meant that every preprocessing step had to be performing in a short time.
- Every prediction on a piece of 20 kilometers of river trained about 20,000 models to obtain a good performance.
- To our knowledge, no one had ever tried modelling rivers with Machine Learning before.
- Every point in the river has its own dynamics. This depends on its depth, distance from the center, seasonal influence and bottom composition.
By showing that Machine Learning can deliver accurate prediction, our client wants to proceed and bring this model into its live system.
Also, the insights gathered during the data exploration, will be shown in a live dashboard.
Conclusion: thanks to the data visualisation and the dashboard, both the accuracy of the reported data and the safety for the ships navigating the river increased.