This letter propose a multiple-input long short-term memory (MI-LSTM) for the prediction algorithm of future movement of an object based on the moves that have been made. First, a representative dynamic 3D space movement model based on random walk mobility. The system forecasts the next movement of an object using the previous movement by utilizing LSTM neural network model. Moreover, performance is evaluated by making a comparison of the propose scheme based on system errors with time-series algorithms; gated recurrent unit (GRU) and BidirectionalLSTM (Bi-LSTM). The result obtained proves that the proposed method works effectively as it predicts the next object¡¯s movement state even though in random behaviour, and also maintaining low computing time in the learning process as a main focus.