Time prediction methods in production − a scoping review of the 21st century so far

• Review of methods for predicting job and manufacturing activity durations. • Analysis of 173 papers between 2000–2024. • Analysis of production times, features, prediction methods and prediction error. • State-of-the-art approaches, prediction methods and research designs are discussed. • Research gaps and future research directions are provided. Accurate prediction of production times is a key aspect of industrial engineering, as it directly affects production planning, product cost estimation, and continuous improvement management. In modern society, where customer expectations for low cost, short lead times, and high quality are constantly rising, inaccurate time estimation can lead to poor decision-making, customer dissatisfaction, and loss of market position. Dynamic changes in demand, supply, and product variety, as well as variations in worker skills, machine conditions, and manufacturing system states, make time prediction increasingly important and challenging, despite the growing availability of data. This scoping review of 173 papers published between 2000 and 2024 examines the state of the art in production time prediction and outlines directions for future research. The studies are classified by the type of predicted time, applied methods, manufacturing systems, and reported accuracy. This classification highlights patterns in the use of different prediction methods and provides insight into their respective performance. The analysis finds that machine learning methods are increasingly used for lead time prediction (approx. 10% error) and human-centred activities (approx. 20% error), while deterministic approaches remain relevant for machining operations (approx. 6% error). Despite the long history of production time prediction and several established practices, gaps remain in methodological standards, including inconsistent definitions of production times, varying validation metrics and procedures, and limited attention to data credibility and system integration. Research on practical applications outside highly automated sectors, such as semiconductor or automotive manufacturing, is also limited, particularly for human-centric tasks expected to grow, including maintenance and collaborative robotics.