VEST: automatic feature engineering for forecasting

Time series forecasting is a challenging task with applications in wide range of domains. Auto-regression one the most common approaches to address these problems. Accordingly, observations are modelled by multiple regression using their past lags as predictor variables. We investigate extension auto-regressive processes statistics which summarise recent dynamics time series. The result our research novel framework called VEST, designed perform feature engineering univariate and numeric automatically. proposed approach works three main steps. First, mapped onto different representations. Second, each representation summarised statistical functions. Finally, filter applied for selection. discovered that combining features generated VEST auto-regression significantly improves performance database composed 90 high sampling frequency. However, we also found there no improvements when multi-step or low sample size. publicly available online.