Multiresolution Forecasting for Industrial Applications

The forecasting of univariate time series poses challenges in industrial applications if the seasonality varies. Typically, a non-varying is treated with model based on Fourier theory or aggregation forecasts from multiple resolution levels. If changes time, various wavelet approaches for are proposed promising potential but without accessible software systematic evaluation different models compared to state-of-the-art methods. In contrast, advantage specific multiresolution here convenience swiftly implementation R and Python combined coefficient selection through evolutionary optimization which evaluated four applications: scheduling call center, planning electricity demand, predicting stocks prices. benchmarking out-of-sample resulting cross-validations error measure MASE SMAPE distribution each method dataset estimated visualized mirrored density plot. performs equal better than twelve comparable methods does not require users set parameters contrary prior frameworks. This makes suitable applications.