CPSL 2024 - 6th Conference on Production Systems and Logistics - Analysis of the Relevance of Models, Influencing Factors and the Point in Time of the Forecast on Prediction Quality

Aug 5, 2026

Upload by Ivy

Manufacturing companies using workshop-based organization rely on communicating feasible delivery dates to customers. Large delivery-time buffers reduce economic efficiency and increase finished-goods inventory when orders finish too early. Machine learning enables dynamic, order-related delivery-time forecasting, yet development requires answers on which influencing factors to include, which models yield the best forecast quality, and when forecasting is economically sensible. Existing methods often consider too few process steps, forecast throughput instead of delivery time, and ignore the available information and external factors at the decision point. This paper evaluates how model choice, forecast timing, and included factors affect achievable forecast quality across five real-world use cases.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages17

Reading Time43 min

AccessFree to download

Abstract

Introduction

Delivery reliability vs economic objectives

Forecast quality improvement via ML

Limitations of existing approaches

CONFERENCE ON PRODUCTION SYSTEMS AND LOGISTICS CPSL 2024 6th Conference on Production Systems and Logistics Analysis of the relevance of models, influencing factors and the point in time of the forecast on the prediction quality in order-related delivery time determination using machine learning Alexander Rokoss 1, Lennart Popkes 1, Matthias Schmidt 1 1Institute for Production Technology and Systems (IPTS) / Leuphana University, Lueneburg, Germany Abstract One of the main objectives of manufacturing companies that structure their manufacturing system according to the workshop principle is to meet the delivery dates communicated to the customer. One approach to avoid large delivery time buffers to stabilize liability of communicated delivery dates is to improve the forecasting quality of the initially determined planned delivery dates. In this context, machine learning methods are a promising approach for the dynamic, order-related forecasting of delivery times. In the development process of machine learning based applications for delivery time forecasting companies are challenged by the following questions: which influencing factors must be considered? Which machine learning models generate the best forecast quality? At what point in the production process does the application of machine learning methods for delivery time forecasting make sense from an economic perspective? Existing approaches do not adequately address these questions. In most cases, only few process steps are considered and only throughput times are forecasted instead of delivery times. The information available at the point in time when the delivery time is forecasted is not discussed. The considered input factors influencing the delivery time are reduced to the company's internal supply chain and therefore do not allow for a satisfactory forecast quality of the delivery time. External influencing factors are often not included. Therefore, this paper describes the influence of different machine learning models, different points in time for the forecasting itself and included influencing factors on the achievable forecast quality. The influence is determined by applying machine learning methods on delivery time forecasting to five real-world use cases. Keywords delivery time, workshop manufacturing, machine learning, artificial intelligence, data mining, 1. Introduction Production companies are confronted with increasing performance requirements in global competition. This competitive pressure is leading to an increasing individualization of products [1] . In this context, the increasing number of variants and customer order-specific production often lead to companies organizing their production more flexible and thus producing in a workshop environment [2] . In make-to-order production, order-based production according to the workshop principle often results in production processes not being completed on time. The large number of customer-specific orders results in long and often widely varying throughput times [3] . This variation is exacerbated by the scheduling of rush orders [2] . Delivery reliability can generally be improved by applying delivery time buffers (also: safety time) [4] . However, Stalk and Hout emphasize the importance of short delivery times for the economic success of a production company [5] . A conflict of objectives arises when designing the delivery time buffer. On the one hand, a high delivery time buffer increases compliance with delivery dates to the customer. On the other hand, the DOI: [https://doi.org/10.15488/17732](https://doi.org/10.15488/17732) ISSN: 2701-6277 415 total delivery time and the finished goods inventory increase due to production orders that have been completed too early [4] . The application of high delivery time buffers as the sole measure to ensure high delivery reliability is unsuitable from an economic perspective. To improve delivery reliability. excluding high delivery time

FAQs

Why is improving delivery-time forecasting quality important in workshop-based manufacturing?

Workshop environments often produce orders not being completed on time, creating long and highly variable throughput times. Improving forecast quality helps reduce reliance on large buffers while maintaining delivery-date compliance.

What key development questions does the paper address for ML-based delivery time forecasting?

It focuses on which influencing factors must be considered, which machine learning models generate the best forecast quality, and at what point in the production process ML forecasting is economically sensible.

How do existing approaches limit practical comparability and reproducibility?

They use different forecasting models, different data bases and preprocessing methods, and perform forecasts at different points in time, which hampers consistent evaluation.

Aug 5, 2026

Upload by Ivy