Thermal runaway is one of the most critical safety challenges associated with aged lithium-ion batteries, and its accurate prediction plays an essential role in developing advanced battery management and early-warning systems. In this study, an integrated framework combining multiphysics modeling with a long short-term memory (LSTM) neural network was developed to predict the thermal runaway temperature of an aged 14500 lithium-ion cell. The physics-based model accounted for the simultaneous effects of calendar aging, solid–electrolyte interphase (SEI) growth, metallic lithium plating, furnace temperature, and the convective heat transfer coefficient. The results demonstrated that increasing the furnace temperature from 418 to 458 K increased the maximum cell temperature from 422.12 to 587.87 K, while reducing the time required to reach the maximum temperature from 2316 to 942 s. Furthermore, increasing the convective heat transfer coefficient reduced the intensity of exothermic reactions and improved the thermal stability of the cell. The LSTM model exhibited excellent performance in predicting the time-dependent temperature history, achieving minimum mean absolute error (MAE) and root mean square error (RMSE) values of 0.0001 and 0.0008, respectively, as well as maximum coefficient of determination (R2) and Nash–Sutcliffe efficiency (NSE) values of 0.9990. These findings demonstrate that the proposed framework provides a reliable tool for the real-time prediction of thermal runaway and the development of intelligent battery management systems.