AI, and AI again, today machine learning or deep learning seems to be on everyone’s lips. But what does it (really) bring to the logistic transport sector? Today we present a very concrete usecase on the use of artificial intelligence to improve the accuracy of shipping routes prediction.

 Let’s go back to the basics

 The Safecube platform offers visibility on the journey of goods thanks to the capture of data from trackers and the recovery of data from external sources such as shipping companies or data providers such as AIS. This data is retrieved on a regular basis with varying degrees of accuracy. In this article, we will focus on a road accuracy issue. Depending on the data points collected, the route does not always correspond to reality. The interest of artificial intelligence is to make the routes visible on the platform as realistic as possible. But the aim is to make the information on the location of the goods more reliable and to predict future positions in order to anticipate potential delays.

AI may seem obscure or even esoteric to some. We attempt to explain how its use can be a real asset  for improving the visibility of our customers on their shipments, based on 3 concepts.

 How does it work? 

The neural network concept

Through an algorithm, the artificial neural network allows the computer to learn from new data. The computer with the neural network learns to perform a task by analyzing examples (data history) to learn from. Given known input and output data, the algorithm must pass through as many points as possible to provide the best answer. 

Like in biology, the set of neurons is organised in layers. From one layer to the next, the input signal is propagated to the output. The principle is to look at the output in relation to what was expected and to update the links between the neurons to improve our final result, which will be a prediction from the network.

 For example, if we train the network to estimate a path taken by the boat, we will give it thousands of examples of paths. The algorithm will predict based on the most recurrent examples. The intelligence will have as many neurons as variables (possibilities / examples)

  • The input layer = a set of neurons that carry the input data.
  • The hidden layers (hidden layer 1, hidden layer 2, …). This is where the relationships between the variables will be highlighted.
  • The output layer: this layer represents the final result of our network, its prediction.

The data history concept

In order to recreate a track as close to reality as possible, the computer needs a large number of examples. These examples are recreated using historical data retrieved from previous IoT)bases trackpoints as well as information provided by shipping companies and AIS data. Safecube’s data history is a record of the tracking of our customers’ shipments. We are currently working on the algorithm with several thousand over 5600 shipments and data points. 

The recurrence concept

The choice of the algorithm will be made according to the recurrence of data. The intelligence will evaluate the most recurrence points to make its choice of result. 

Safecube is gradually integrating AI into the route calculation to be as accurate as possible. The objective of this work is to create an agnostic algorithm to make it usable on all our solutions: IoT and 100% digital