In radiation therapy, precise dose distribution is essential for minimizing damage to normal tissues. Gafchromic EBT3 film is widely used to assure the quality of two-dimensional dosimetry but requires frequent recalibrations due to changes in sensitivity over time. This study presents a new calibration method using a Keras-based generalized additive neural network (GANN) to address film aging. EBT3 films from four lots were calibrated with a 6 MV photon beam and scanned on Epson scanners. The GANN method achieved percentage differences between calibrated and delivered doses within 5%, comparable to traditional recalibration methods, with an overall uncertainty of approximately 2%. It demonstrated improved stability and accuracy, significantly reducing the need for frequent recalibrations and providing a robust solution for long-term film dosimetry.