|
Hi, everyone Recently, I encounter a problem about time series analysis with constraints. Specifically, given the following data for n time points from t_1 to t_N: feature: f_{t_1}, f_{t_2}, ..., f_{t_N} label: l_{t_1}, l_{t_2}, ..., l_{t_N} constraint: l_{t_n} in {1, 2, 3, 4, 5}, and l_{t_i} <= l_{t_j} if t_i < t_j I want to construct a model to predict l_{t_{N + 1}} from the above data and f_{t_{N + 1}}. Does someone know how to build a good model for this time series analysis? Or give me some clues and discusses. Thanks in advance. |
|
The model you choose will depend on the distribution of the time series data. Suggest you make a scatter plot to ascertain the trend if any. If there is any significant upward or down trend then your options for forecasting are Linear, Quadratic, Exponential, Holt-Winters or any auto aggressive models ( assumed that the time is annual); otherwise least squares forecasting. If there is no trend use Exponential smoothing or moving averages |