TimeSeriesDecompositionReport<T>
Models & Types · AiDotNet.DecompositionMethods.TimeSeriesDecomposition
The time-series decomposition audit produced by a configured decomposition: not just the components, but a grade of how good the decomposition is (trend/seasonal strength, leftover residual structure, additive reconstruction fidelity) and a measure of whether decomposition-based forecasting would actually help.
How It Works
Most libraries hand back trend/seasonal/residual and stop. This report grades the split: trend and seasonal strength (Wang/Hyndman variance-ratio measures, 0..1), whether the residual still carries autocorrelated structure the decomposition missed, and whether the components add back to the original series. It also forecasts a held-out tail through the decomposition (trend extrapolation + seasonal-naive repeat) versus a random-walk baseline, so the reported skill says whether decomposing the series is worth it.
Constructors
| Constructor | Summary |
|---|---|
TimeSeriesDecompositionReport() |
Properties
| Property | Summary |
|---|---|
DecompositionAnalysisSource AnalyzedSeries { get; set; } |
Which series was analyzed — the model’s training target, or the configured instance’s own series. |
IReadOnlyList<DecompositionComponentType> ComponentsPresent { get; set; } |
The decomposition components that were present and used. |
bool ForecastEvaluated { get; set; } |
Whether the held-out forecasting comparison was evaluated (enough points and components). |
bool ReconstructionAvailable { get; set; } |
Whether an additive reconstruction was computable (trend and residual components were present). |
bool ResidualHasStructure { get; set; } |
Whether the residual still carries structure the decomposition failed to capture (its lag-1 autocorrelation exceeds the ~2/sqrt(n) white-noise band). |
double DecompositionForecastRmse { get; set; } |
RMSE of the decomposition-based forecast (trend extrapolation + seasonal-naive) on the held-out tail. |
double ForecastSkill { get; set; } |
Forecast skill of decomposition over the naive baseline: 1 minus decomposition RMSE over naive RMSE. |
double NaiveForecastRmse { get; set; } |
RMSE of the random-walk (last-value carry-forward) baseline on the same held-out tail. |
double ReconstructionError { get; set; } |
Root-mean-square error of the additive reconstruction (trend + seasonal + residual) versus the series. |
double ResidualAutocorrelation { get; set; } |
Lag-1 autocorrelation of the residual — near 0 means the residual looks like white noise. |
double SeasonalStrength { get; set; } |
Seasonal strength F_S in 0..1 (Wang/Hyndman): 1 minus the variance of the residual over the variance of the deseasonalized-plus-residual. |
double TrendStrength { get; set; } |
Trend strength F_T in 0..1 (Wang/Hyndman): 1 minus the variance of the residual over the variance of the detrended-plus-residual. |
int ForecastHorizon { get; set; } |
The number of tail points held out for the forecasting-value comparison. |
int SeriesLength { get; set; } |
The length of the analyzed series. |
string MethodName { get; set; } |
The configured decomposition method’s type name. |