Development of an AI-based system for forecasting the transit times of parcel deliveries
Project duration: 8 months
Brief description
The aim of the project is to develop an AI-based system for the automated forecasting of parcel deliveries. For a predefined period of time and client of the customer, this prototype determines the expected delivery date of the consignments (parcels) every calendar day depending on the time of picking and the delivery area.
Supplement
PTA advises on the conception and development of a forecasting system which determines the most probable delivery date of a parcel delivery on the basis of machine learning and historical data. Within the framework of so-called feature engineering, PTA supports the customer by means of explorative data analysis and suitable statistical methods to identify features relevant to the forecast. On the basis of historical data previously collected with SQL scripts, different models are evaluated against each other or subjected to benchmarking using so-called supervised learning procedures. The procedure with the lowest prediction error is implemented with Python (scikit-learn) within the framework of an ETL pipeline. Deployed in the cloud, the pipeline is executed automatically on a daily basis. The results of the pipeline execution (forecasts) are reported to the client's customer via email.
Subject description
The overall project is a cooperation between CEP service providers and an online retailer (remote retailer). The goal of the online retailer is to offer his end customers a special customer experience: He would like to give them a reliable delivery date already at the time of enquiry and order. The online retailer takes care of procurement, quality inspection, storage and picking of the shipments. The CEP service provider collects the shipments and is responsible for the delivery process. In the CEP service provider project described here, a reliable statement is determined for the duration of the delivery process for the end customer's order. Especially in times of increased shipment volume and the resulting increased capacity utilisation of the logistics network, delays in the delivery of shipments can occur. With the help of the developed forecasting system, probabilities for transit times of consignments are determined on the basis of machine learning methods and historical data.