A Survey on Long-Short-Term-Memory Networks and its Applications

Long Short-Term Memory (LSTM) Network, a category of Recurrent Neural Networks (RNN) is effective for time series data as it maintains information for a longer time in the memory and resolves gradient problems. It is proficient in learning the past stages information and utilizing the same in future predictions. Two major applications Stock price prediction and financial fraud detection are considered in the paper for evaluating the performance of LSTM .Experimental results reveal that, for stock price prediction, the model achieves least MAPE 1.603 for the Bank of Baroda dataset and Minimum MAE value of 1.421 for the Punjab National Bank dataset. In fraud detection system, LSTM achieves the performance accuracy of 99.98% in a short time using Adam optimizer.