Using predictive models for data-driven optimization of production lines is a trending topic in manufacturing. A challenge is to support the visualization and management of data and arising uncertainties in various phases. Modern production lines can contain thousands of sensors to collect readings every few milliseconds. When creating predictive models for optimization from this data, its sheer amount not only poses challenges to data storage and management, but also during the different phases of data preparation, modeling, decision making, and executing optimizations. Our research aims at a better support of all these phases with data visualization, with new strategies for the management of uncertainty. Predictive models help operators to identify problems and the degree of wear of tools or parts in production lines. However, data analysts, supervisors, and service technicians must be aware that these models contain uncertainties and the recommendations made by these models should be critically reflected before acceptance. One way to mitigate the negative impact of uncertainties is to use visualization as tool for analyzing the potentials and limits of prediction models. Therefore, we propose a process model for the different activities of optimization in a smart production environment and discuss challenges for the visualization of data and uncertainty in each step.