Predictions are a major part of our AI data pipeline. Predictions learn from past carrier behavior, and 3rd party market data, in order to predict market price and carrier-specific preferences or reactions. There are many ways to make market price or carrier-level predictions, but the diagram below explains one of our more effective techniques for market rates.
We’ve identified 80 key features of the transport task, such as distance driven, timing of pickup etc. The platform automatically generates prediction trees to relate each feature to the others in if-then evaluations. Each tree has roughly 6k branches and ends with a price predicting formula for each circumstance. An ensemble method allows around 200 pricing trees to vote on the predicted price of a new shipment. Learn about how prediction fits into our AI pipeline here.