The increasing urgency to ensure nuclear reactor safety under extreme conditions, such as Station Blackout (SBO), has led to the exploration of advanced modeling techniques. This article presents an LSTM-based machine learning model for the prediction of key thermal-hydraulic parameters under SBO conditions in an AP1000 Pressurized Water Reactor (PWR) based on dataset generated by RELAP5. RELAP5 simulated 120 SBO transients with different time delays in engineered safety feature (ESF) actuation and produced time-series data to train and test the LSTM model. Feature selection by correlation heatmaps cut down input variables from 34 to 17, enhancing computational efficiency. The model was trained for the prediction of Peak Cladding Temperature (PCT) and Hot Leg Temperature with a sliding window method and optimized neurons, layers, and time step parameters. Performance metrics such as MAE, MAPE, and RMSE exhibited excellent prediction performance, especially for Hot Leg Temperature with RMSE = 2.55 and MAPE = 0.34 %. While being marginally more volatile, the model also had a very low MAPE of 2.05 % for PCT prediction. The findings validate the capability of utilizing LSTM networks for accurate real-time prediction of safety-critical parameters with remarkable improvement in risk-informed decision-making and accident management. © 2025 Elsevier Ltd. All rights are reserved, including those for text and data mining, AI training, and similar technologies.