The reliability of Insulated Gate Bipolar Transistor (IGBT) modules is critical for long-term durability and operational stability in power electronic systems. This paper presents a real-time degradation-monitoring and lifetime-prediction framework for IGBT modules using accelerated power-cycling tests under realistic inverter conditions. A two-level SVPWM inverter is used, with electrically driven current loading and developed induction heating, along with an active water-cooling system, to create precise junction-temperature swings to investigate thermal fatigue. Continuous acquisition of on-state voltage and temperature data is performed throughout the aging of six IGBT modules under two junction-swing conditions. The degradation behavior is examined through experimentally observed trends and used to develop a machine-learning model based on Random Forest (RF) to predict the remaining useful lifetime (RUL) utilizing the extracted degradation indicators. A comparative analysis with a Support Vector Regression (SVR) and Gaussian Process Regression (GPR) models is performed to validate the RF model’s effectiveness, confirming that the RF model has greater accuracy, robustness, and computational efficiency with limited datasets. The proposed framework thus integrates experimental testing, and data-driven prediction, attaining lifetime-prediction errors under 4% while utilizing only 20–60% of the lifetime data, offering a precise and scalable method for assessing inverter-level IGBT reliability.