Distributed Stream Processing (DSP) systems enable processing large streams\nof continuous data to produce results in near to real time. They are an\nessential part of many data-intensive applications and analytics platforms. The\nrate at which events arrive at DSP systems can vary considerably over time,\nwhich may be due to trends, cyclic, and seasonal patterns within the data\nstreams. A priori knowledge of incoming workloads enables proactive approaches\nto resource management and optimization tasks such as dynamic scaling, live\nmigration of resources, and the tuning of configuration parameters during\nrun-times, thus leading to a potentially better Quality of Service.\n In this paper we conduct a comprehensive evaluation of different load\nprediction techniques for DSP jobs. We identify three use-cases and formulate\nrequirements for making load predictions specific to DSP jobs. Automatically\noptimized classical and Deep Learning methods are being evaluated on nine\ndifferent datasets from typical DSP domains, i.e. the IoT, Web 2.0, and cluster\nmonitoring. We compare model performance with respect to overall accuracy and\ntraining duration. Our results show that the Deep Learning methods provide the\nmost accurate load predictions for the majority of the evaluated datasets.\n