This study employs a deep learning approach to develop a Long Short-Term Memory (LSTM) neural network model for predicting the seismic response of self-centering braced steel frame structures. The model establishes a nonlinear relationship between input seismic acceleration and output structural drift responses. The impact of data window size on predictive accuracy is examined. Results indicate that the LSTM model demonstrates robust predictive capabilities. Increasing the data window size enhances prediction efficiency while maintaining accuracy for both roof drift and inter-story drift, with minimal influence on the peak phase distribution of the predicted responses.