Veeroute takes advantage of best practices in artificial intelligence.
It uses boosting algorithms for machine learning to select the configurations
that best match the characteristics of your input dataset.
The Veeroute optimization engine uses configurations - sets of algorithms and
routing models it applies in a specific sequence - to solve combinatorial problems
such as last mile delivery optimization. The input dataset which describes the
system will prompt the optimization engine to use one of several possible
configurations to perform the calculations. This means Veeroute must quickly
analyze the input dataset and identify the best configuration.
Veeroute does this by using a boosting algorithm with multiple weak learners.
Represented in the picture by decision trees, a weak learner is a simple
model/decision tree that performs slightly better than a random choice.
A model uses the inputs from these weak learners to decide which configurations
the optimization engine should use to solve the problem.
Each weak learner’s input may have a different weight in the final decision
that represents the importance of it's input.
- The best configurations
The configurations suggested by AI are applied to solve the defined combinatorial problem. To avoid misclassification, we train the weak learners on the set of datasets which we continually adjust and extend based on the datasets we receive from our customers.