Location:Campus Albano, Lecture room 25, house 4, level 2
Time Series Features and LSTM Forecasting
Abstract
Long Short-Term Memory (LSTM) networks are benchmark deep learning models for time series forecasting, yet the factors driving their performance remain unclear. We extract time series features from the M4 dataset and analyze how they influence LSTM forecasting error using random forest importance and regression models. Results show that series length per forecast horizon, skewness, autocorrelation structure, and stationarity strongly predict LSTM performance, while nonlinearity and persistence have little impact. These findings clarify when LSTMs perform well and offer practical guidance for their application in forecasting tasks.
