1: \begin{abstract}

2: %Conventional change point methods for calibrating discrete event simulation models use data-driven approaches. While some change points are external to the process based system (i.e., arrival patterns), others (e.g., resource schedules, control policy changes) are internal and could be discovered by leveraging process knowledge. The proposed method augments these data-driven approaches through the integrated use of simulation models and machine learning methods that enable the discovery of process drivers to improve change point detection accuracy. Proposed method is shown to leverage process knowledge in improving the data-driven change point detection to pinpoint the precise changes in the process-based system's internal drivers. Three points are important to better implement this method in practice: 1) In the presence of a large number of features, it could be more efficient to use feature selection techniques to identify a subset of time series features that are significant to be used in the model. 2) Practitioners may either use arrival data as an input to their model, or simulate arrival through a specific distribution. Former method would accelerate convergence for a particular scenario, whereas the latter would enable the reuse of the results of the simulation models for similar analyses. 3) Since the data driven methods provide an initial solution for our method, it is important to use a data driven method that works effectively in order to improve computational efficiency. We will research multiple extensions of this study in the future.  

3: 

4: %DD-CPD method is used to provide a coarse estimate, it is important to use a DD-CPD method that is accurate to improve solution performance

5: 

6: 

7: %feature selection

8: %if time series features are high -> feature selection methods could be used such as 

9: %feature importance metric (using machine learning such as random forest to order features)

10: %\end{abstract}

11: