The current study explores deep learning techniques to investigate ocean current velocity and predictability. We utilize the long short-term memory (LSTM) deep learning network to analyze ocean current data. Here, a time series of ocean currents data at one location obtained from the Indian Space Research Organisation’s (ISRO) Meteorological and Oceanographic Satellite Data Archival Centre (MOSDAC) is used. We demonstrate that the LSTM can predict the current speed (V) as well as the two velocity elements, zonal velocity (u) and meridional velocity (v). The performances of three LSTM network variants, viz., vanilla LSTM, stacked LSTM, and bidirectional LSTM, are compared. The evaluation metrics used here are (i) root mean squared error (RMSE), (ii) mean absolute error (MAE), and (iii) mean of prediction (MoP). We also investigate how the number of epochs affects the predictions. Our findings have numerous relevant applications, including weather prediction and predicting tidal energy variance. They have a high potential for use in various practical problems in physical oceanography.