In order to ensure adherence to schedules, knowledge of planned lead times (LT) is crucial for success. In practice, however, rigid planning methods are often used which cannot adequately reflect constantly changing environmental influences (e.g. fluctuations in the daily workload). Particularly in job shop production, precise planning of LT is difficult to implement. This paper therefore examines whether existing machine learning (ML) approaches, in particular supervised learning methods, in production planning can support LT scheduling in job shop production to generate added value. The paper enhances existing research by comparing deep artificial neural networks with ensemble methods (e.g. random forest, boosting decision trees). The applied approach bases on the Cross Industry Standard Process for Data Mining (CRISP-DM), which was created by a consortium of companies. Finally, the evaluation through an exemplary job shop production shows that the present work contributes to mastering the planned LT. In particular, the ML model, boosting decision trees and deep artificial neural networks show significant improvements in planning quality. This practical reference has not yet been addressed comprehensively in the literature.