Forecasting machine performance check output using Holt-Winters approach

Background: Machine Performance Check (MPC) is an automated TrueBeam quality control (QC) tool used to verify beam output, isocenter, and uniformity. The aim of this study was to build an MPC output variation time series modeled on the Holt-Winters method over thirty days. Methods: After AAPM TG-51 and baseline data were established for the Edge TrueBeam, daily MPC output data were gathered and analyzed through a Holt-Winters (additive and multiplicative) method. The model's performance was assessed via three standard error measures: the mean squared error (MSE), the mean absolute percentage error (MAPE), and the mean absolute deviation (MAE). The aim was achieved using a nonlinear multistart solver on the Excel platform. Results: The results showed that MPC output variation forecasting is energy and model dependent. Both additive and multiplicative Holt-Winters methods were suitable for the analysis. The performance metrics MSE, MAPE, and MAD were found to be well within acceptable limits. Conclusions: A Holt-Winters model was able to accurately forecast the MPC output variation.