2. Materials and Methods
The artificial intelligence (AI) discussed here was developed based on a big dataset derived from an existing drive-through mass vaccination simulation created by the authors [32]. In this section, we briefly explain the drive-through simulation tool and then provide some details about our machine learning model.

2.1. The Drive-Through Simulation
The drive-through model used in this study is a hybrid model consisting of a discrete event and an agent-based simulation. The model contains several agents including passengers, staff, and cars. The physical layout of the drive-through (Figure 1) can be extended up to ten lanes with a maximum area of 20,000 (150 m by 130 m). Passengers entering the drive-through are screened at the screening station to ensure that they meet the vaccination criteria and receive the necessary handout materials. Cars and passengers passed the screening phase move to one of the open lanes for registration, vaccination, and recovery, which can be determined in part by the availability of High Occupancy Vehicle Lane and Pre Registered Lanes. Vaccination will be conducted by professional immunization staff at the vaccination stations. Cars move to one of the available parking spaces located after the vaccination stations. Individuals experiencing complications after vaccination can be taken care of by the relevant staff. Each station is supported by up to four kiosks.
The discrete event model of the drive-through simulation generates the service process for each lane from registration to recovery (Figure 2). Start and end time in each station is measured through time blocks (for example, timer and timeRe refer to the start and end time for registration respectively).
A sample of our drive-through simulation experiment results is presented here. Table 1 shows the parameters setting for the experiment and Figure 3 presents the output of the experiment. The values presented in Figure 3 are the average of 50 iterations of this experiment. According to these results, a total of 1933 cars with 5804 passengers will be processed in the drive-through mass vaccination during a day with three shifts of 8 h each.

2.2. The AI Model
The high fidelity AnyLogic model suffers at inference time, due to the computation cost of running the simulation. A single simulation run can take up to 90 s, which may not be efficient in any real time analysis tasks. Due to the stochastic nature of the drive-through simulation, a significant number of simulations runs (using Monte Carlo method) is needed for each parameter setting that requires more simulation time. In order to alleviate this, we attempt to train a neural network to predict the outputs of the simulation based on the model parameters.
As is often the case with training neural networks, the problem is rooted in gathering large amounts of data on which to train. For this, we make use of the AnyLogic parallelized computation ability, to simulate large batches of simulation runs at the same time. In doing so, we strategically sample across a large range of parameters, near the domain of interest required of real-world scenarios. Parameter simulation ranges can be found in Table 2.
We chose to sample all variables uniformly, as we do not have any distribution knowledge of the real-world application of the model. Furthermore, by sampling in such an unbiased fashion, we reduce the chances of the network overfitting based on some intrinsic property of our training set. Some conditional restrictions are built regarding the binary variables, as noted in Table 1. Altogether, we generated about 125 k simulation samples, providing us with a dense enough training domain.