AutoPas is a high-performance, auto-tuned particle simulation software library for many-body particle systems, capable of dynamically switching between algorithms and data structures to achieve optimal performance during the simulation. This thesis explores a decision tree-based tuning strategy for AutoPas, allowing users to guide the tuning process by specifying custom decision tree or random forest models, which can be used to efficiently conclude the tuning phase and determine the best configuration from the search space. Efficient tuning strategies are crucial, as they allow for discarding poor configurations, thus enabling the simulation to perform optimally.

We demonstrate that an approach where we train a decision tree or a random forest model based on data collected from already existing Full Search tuning strategy can significantly outperform existing tuning strategies on specific benchmarks by drastically reducing the tuning time and in turn, the total computation time.

The decision tree-based tuning strategy can drastically reduce the time spent during the tuning phases while achieving comparable tuning results; therefore, it can be an alternative candidate to the current tuning strategies.