Robust Time Series Forecasting via Time Differencing and Stacking

Robust Time Series Forecasting via Time Differencing and Stacking (IEEE SMART GENCON)

Designing a stationarity-enforced stacking architecture for financial/metric forecasting.

• Mathematical Rigor: Resolved non-stationarity in raw series by applying first-order Time Differencing to eliminate trend and seasonality components.
• Feature Engineering: Built a robust input space using lag features ($t-1$, $t-2$, $t-3$) alongside technical indicators.
• Architecture: Formulated a stacked ensemble utilizing diverse linear and non-linear base estimators combined with a meta-regressor, reducing forecasting variance across noisy datasets.
• Publication: IEEE SMART GENCON