Zaman serilerinin karmaşık tam sayılı programlama ile parçalarına ayrıştırılması

Decomposing time series into seasonality, trend, and remainder reveals underlying insights to be used in forecasting and anomaly detection. Although there are several decomposition methods, no method guarantees all of the following issues are addressed: i) smoothness of trend and the rigid structure of seasonality, ii) shifts in trend, iii) long seasonality periods, iv) multi-seasonality, and v) robustness on outliers. In this study, we propose a mixed integer programming model to address all of these issues. Experiments on different synthetic problem sets present the effectiveness of the proposed algorithm, providing benchmark results against the robust seasonal trend decomposition algorithm.