The uncertainty estimation of feature-based forecast combinations

Forecasting is an indispensable element of operational research (OR) and important aid to planning. The accurate estimation the forecast uncertainty facilitates several operations management activities, predominantly in supporting decisions inventory supply chain effectively setting safety stocks. In this paper, we introduce a feature-based framework, which links relationship between time series features interval forecasting performance into providing reliable forecasts. We propose optimal threshold ratio searching algorithm new weight determination mechanism for selecting appropriate subset models assigning combination weights each tailored observed features. evaluate our approach using large set from M4 competition. Our experiments show that significantly outperforms wide range benchmark models, both terms point forecasts as well prediction intervals.