Abstract Floating offshore wind turbines (FOWTs) are emerging as a promising renewable energy solution, tapping into the vast wind resources in deep-sea locations. Accurate predictions of heave, pitch, sway, roll, yaw, and surge are essential to ensuring the structural integrity and power generation efficiency of FOWTs. Traditional methods relying on potential flow theory to predict FOWTs’ hydrodynamic responses involve simplifications that may not capture real-world complexities. Although computational fluid dynamics (CFD) simulations offer accuracy, they are computationally expensive. This study explores the use of Bi-directional long-short-term memory (Bi-LSTM) neural networks as an alternative to address computational challenges in predicting FOWTs’ hydrodynamic responses. These networks excel at handling time series data, capturing temporal correlations crucial for understanding FOWT floater motion dynamics. They also effectively capture bidirectional dependencies, enhancing their suitability for this application. The model considers various input parameters, including wave amplitudes, periods, steepness, directions, and floater design characteristics such as stifness, pre-tension, and the free length of mooring lines. The Bi-LSTM model is trained using a subset of the CFD dataset generated for the scaled OC5 semi-submersible platform with mooring, reserving the remainder for validation and testing. The comparison between simulated and predicted motion demonstrated a strong correlation in surge but noticeable discrepancies in heave and pitch. FFT analysis highlighted consistent frequency components but variations in magnitude, suggesting areas for model sensitivity improvement and further investigation into wave interaction dynamics. The correlation coefficients revealed a strong positive correlation for the surge motion, while the heave and pitch motions showed slightly lower correlation coefficients, suggesting better performance in predicting the surge.