ETA Prediction using Feed Forward Neural Network

Introduction

Predictive eta, or estimated time of arrival, is a crucial metric for transportation companies and logistics providers. Accurate eta predictions help transportation companies plan their operations, reduce delays, and improve customer satisfaction. In this blog post, we will discuss how to use Neo4j as a database and feedforward neural networks to predict the eta of multimodal transport, taking into account various factors such as route distance, vehicle/vessel speed, intermediate stops, weather conditions, refuelling time, traffic, hours of service, and mandatory rest periods for truck drivers.

Neo4j as a Graph Database

Neo4j is a popular graph database that is well-suited for modelling and querying complex data relationships. Neo4j can be used to store and manage data related to the various transportation modes, routes, vehicles/vessels involved in a multimodal transport operation and the cost involved with each leg of the journey.

For the Shipment Planner, Neo4j is used to store information about the different transportation modes (e.g. truck, train, ship), the routes between origin and destination, the vehicle/vessel characteristics (e.g. speed, capacity), the intermediate stops and their duration, along the route. Using this data, Neo4j can then be used to query for the various possible vehicle/vessel/ports to be taken and the cost involved to reach the final destination. For example, we can use Neo4j to calculate the total distance, time and cost needed for each transportation mode, taking into account the vehicle/vessel speed, the intermediate stops and cost associated with each leg of the journey.

Real Time Tracking API

LognAi uses advanced technology to track shipments in real time, collecting thousands of data points throughout the journey. This includes timestamps for planned and actual arrival and departure times, as well as information about the vehicles or vessels involved. With these points with data points linked to vehicles/vessels, we were able to use this data for training our Feed Forward Neural Network.

Incidents API

In addition to the tracking data, we also use an incidents API to monitor and manage any issues that may arise during transportation. This could include accidents, delays, port congestions, or other factors that could affect the ETA of a shipment. The information from the incidents API is fed into the neural network as input, allowing it to account for these factors when making predictions

ETA Prediction with Feedforward Neural Networks

Once we have the data in Neo4j along with historical Real Time Tracking data and Incidents Data, we can use feedforward neural networks to predict the eta of a multimodal transport operation. Feedforward neural networks are a type of artificial neural network that can learn to predict output values based on a given set of input values. In the context of predictive eta, we can use feedforward neural networks to predict the eta of a multimodal transport operation based on the various factors that affect the eta, such as route distance, vehicle/vessel speed, intermediate stops, weather conditions, refuelling time, traffic, hours of service, and mandatory rest periods for truck drivers.

To train a feedforward neural network, we first need to gather a large dataset of multimodal transport operations, including the input factors (e.g. route distance, vehicle/vessel speed, intermediate stops, weather conditions, etc.) and the corresponding eta. We can then use this dataset to train the feedforward neural network to accurately predict the eta of a multimodal transport operation based on the input factors. Once the feedforward neural network is trained, we can use it to make eta predictions for new multimodal transport operations. Given a set of input factors (e.g. route distance, vehicle/vessel speed, intermediate stops, weather conditions, etc.), the feedforward neural network will predict the eta of the transport operation. These predictions can then be used by transportation companies and logistics providers to plan their operations and improve the accuracy of their eta estimates

The following is a high-level block diagram of our AWS setup.

Conclusion

By using advanced machine learning techniques, we are able to provide more accurate and reliable ETA predictions for our clients, ensuring that their shipments arrive on time and without unexpected delays. This is just one example of how our company is using cutting-edge technology to improve our services and provide the best possible experience for our customers.