LSTM (long short term memory) as the name suggests with huge data sets, can remember certain patterns for longer time frame, whereas the traditional time series prediction involves using the points in preceding the end point in a defined way. With this, LSTMs are much more complex to build and hence a better choice for a prediction on point that is far from the latest available point in the dataset. However, for near term predictions, traditional forecasting techniques are more accurate and simpler to build.