The Advanced Photon Source (APS) at Argonne National Laboratory is a user research facility funded by the U.S. Department of Energy dedicated to research of matter using X-rays. XRF-Maps is the software program used by the Microscopy Group beamlines at the APS to fit the X-ray fluorescence spectra from microscopy experiments to determine the concentration and location of elements within a sample. The fitting routine used by XRF-Maps requires the user to input several parameters to determine the heights and shapes of the peaks. The optimal values of the fitting parameters are not precisely known by the researchers and are tedious to determine. The purpose of this study was to develop a method of quickly suggesting the values of the fitting parameters to the researchers. Three different types of models were compared: random forest, XGBoost, and neural networks. The models were evaluated on their ability to predict the values of the fitting parameters compared to the values chosen in the past by the beamline users. Next, the concentration of elements output by XRF-Maps based on the predicted fitting parameters was compared to the elements given from the ground truth fitting parameters. Finally, the spectra were fitted using the predicted parameters. Overall, the random forest and XGBoost models showed promise for their ability to suggest viable values for several of the fitting parameters; however, performance was varied across the parameters as well as for different element types.