Class ARIMAOptions<T>
Configuration options for the ARIMA (AutoRegressive Integrated Moving Average) time series forecasting model.
public class ARIMAOptions<T> : TimeSeriesRegressionOptions<T>Type Parameters
TThe data type of the time series values.
- Inheritance
-
ARIMAOptions<T>
- Inherited Members
Remarks
ARIMA is a statistical model used for analyzing and forecasting time series data. It combines three components: AutoRegressive (AR), Integrated (I), and Moving Average (MA) to model time series data that exhibits non-stationarity.
For Beginners: ARIMA is a popular method for predicting future values in a time series (data collected over time, like daily temperatures or monthly sales). It works by combining three techniques: looking at how past values influence future ones (AutoRegressive), removing trends by taking differences between consecutive values (Integrated), and accounting for the impact of past prediction errors (Moving Average). Think of it like predicting tomorrow's weather by considering today's weather, the recent trend of warming or cooling, and how accurate previous forecasts have been.
Properties
AnomalyThresholdSigma
Gets or sets the number of standard deviations from the mean residual to use as the anomaly threshold.
public double AnomalyThresholdSigma { get; set; }Property Value
- double
The anomaly threshold in standard deviations, defaulting to 3.0.
Remarks
This parameter controls how sensitive the anomaly detection is. A value of 3.0 means that data points with residuals more than 3 standard deviations from the mean are flagged as anomalies. Lower values (like 2.0) will flag more points as anomalies, while higher values (like 4.0) will only flag the most extreme deviations.
For Beginners: This controls how unusual a value needs to be before it's flagged as an anomaly. Think of it like a "weirdness threshold": - A value of 2.0 is lenient - flags anything moderately unusual (about 5% of normal values might be flagged) - A value of 3.0 (default) is standard - flags clearly unusual values (about 0.3% of normal values) - A value of 4.0 is strict - only flags extreme outliers (about 0.01% of normal values)
If you're seeing too many false positives (normal values flagged as anomalies), increase this value. If you're missing real anomalies, decrease this value.
D
Gets or sets the order of differencing (Integration).
public int D { get; set; }Property Value
- int
The differencing order, defaulting to 1.
Remarks
The differencing order (D) specifies how many times the data is differenced to achieve stationarity. Differencing helps remove trends and seasonal patterns from the data.
For Beginners: This determines how many times the model subtracts consecutive values to remove trends. With the default value of 1, the model works with the changes between consecutive values rather than the raw values themselves. For example, instead of using temperatures like [70°F, 72°F, 75°F], it would use the differences [2°F, 3°F]. This helps the model focus on how values are changing rather than their absolute values, which is useful when data has an upward or downward trend. Think of it like focusing on how much warmer or cooler it gets each day rather than the actual temperature.
EnableAnomalyDetection
Gets or sets a value indicating whether to enable anomaly detection during and after training.
public bool EnableAnomalyDetection { get; set; }Property Value
- bool
True to enable anomaly detection, defaulting to false.
Remarks
When enabled, the model will compute prediction residuals during training and use them to establish an anomaly threshold based on statistical properties of the residuals.
For Beginners: Setting this to true tells the model to not just predict values, but also identify unusual data points (anomalies). An anomaly is a value that deviates significantly from what the model would expect based on the historical patterns it has learned. For example, a sudden spike or drop that doesn't fit the usual pattern would be flagged as an anomaly.
FitIntercept
Gets or sets whether to include an intercept (constant term) in the model.
public bool FitIntercept { get; set; }Property Value
- bool
True to include an intercept, defaulting to true.
Remarks
When true, the model includes a constant term that represents the base level of the time series when all other factors are zero.
For Beginners: This determines whether the model includes a "starting point" or baseline value. With the default value of true, the model can adjust this baseline up or down to better fit the data. Think of it like setting a thermostat - the intercept is like the base temperature setting, and the other parameters make adjustments from there. Having an intercept usually improves predictions, especially when your data doesn't naturally center around zero.
LearningRate
Gets or sets the learning rate for the optimization algorithm.
public double LearningRate { get; set; }Property Value
- double
The learning rate, defaulting to 0.01.
Remarks
The learning rate controls how quickly the model parameters are updated during training. A smaller value leads to slower but potentially more precise convergence.
For Beginners: This controls how quickly the model adjusts its predictions based on errors. The default value of 0.01 means the model makes small, cautious adjustments. Think of it like turning a steering wheel - a small learning rate makes tiny adjustments (good for fine-tuning), while a larger value makes bigger adjustments (which might help learn faster but could overshoot the optimal solution). For most cases, the default small value works well because it helps the model find more accurate predictions, even if it takes a bit longer.
MaxIterations
Gets or sets the maximum number of iterations for the optimization algorithm.
public int MaxIterations { get; set; }Property Value
- int
The maximum number of iterations, defaulting to 1000.
Remarks
This parameter limits how long the model will train before stopping, preventing excessive computation time. The algorithm will stop either when it converges or when it reaches this number of iterations.
For Beginners: This is simply a safety limit on how long the model will try to improve itself. The default value of 1000 means the model will make at most 1000 attempts to refine its predictions before stopping. Think of it like telling someone they can have up to 1000 tries to solve a puzzle - they might solve it sooner, but they won't keep trying forever if they're struggling to make progress.
P
Gets or sets the order of the AutoRegressive (AR) component.
public int P { get; set; }Property Value
- int
The AR order, defaulting to 1.
Remarks
The AR order (P) specifies how many previous time steps are used to predict the current value. For example, P=1 means the model uses only the immediately preceding value.
For Beginners: This controls how many past data points the model looks at to make predictions. With the default value of 1, the model only considers what happened in the previous time period. If you set it to 2, it would look at the last two time periods, and so on. Think of it like predicting tomorrow's temperature based on today's temperature (P=1) versus considering both today and yesterday (P=2). Higher values can capture more complex patterns but might make the model unnecessarily complicated for simple data.
Q
Gets or sets the order of the Moving Average (MA) component.
public int Q { get; set; }Property Value
- int
The MA order, defaulting to 1.
Remarks
The MA order (Q) specifies how many previous forecast errors are used in the model. For example, Q=1 means the model incorporates the error from the previous prediction.
For Beginners: This controls how many past prediction errors the model considers. With the default value of 1, the model adjusts based on how wrong its most recent prediction was. Think of it like a weather forecaster who not only looks at yesterday's weather but also considers how wrong their previous forecast was, learning from their mistakes. If yesterday's prediction was too high, they might adjust today's prediction downward a bit to compensate.
SeasonalD
Gets or sets the seasonal differencing order for seasonal ARIMA models.
public int SeasonalD { get; set; }Property Value
- int
The seasonal differencing order, defaulting to 1.
Remarks
The seasonal differencing order specifies how many times the data is differenced at the seasonal level to remove seasonal trends.
For Beginners: This is like the regular D value, but for seasonal differences. With the default value of 1, the model compares each value to the same period in the previous season. For example, it might look at the difference between this December's sales and last December's sales. This helps remove predictable seasonal patterns, allowing the model to focus on other factors affecting the data.
SeasonalP
Gets or sets the seasonal AutoRegressive order for seasonal ARIMA models.
public int SeasonalP { get; set; }Property Value
- int
The seasonal AR order, defaulting to 1.
Remarks
The seasonal AR order specifies how many previous seasonal time steps are used to predict the current value. This captures patterns that repeat at regular intervals (e.g., yearly, monthly).
For Beginners: This is similar to the regular P value, but for seasonal patterns. It determines how many past seasons influence the current prediction. With the default value of 1, the model considers what happened in the same period last season. For example, when predicting December sales, it would consider last December's sales. This helps capture yearly patterns like holiday shopping spikes or seasonal temperature changes.
SeasonalQ
Gets or sets the seasonal Moving Average order for seasonal ARIMA models.
public int SeasonalQ { get; set; }Property Value
- int
The seasonal MA order, defaulting to 1.
Remarks
The seasonal MA order specifies how many previous seasonal forecast errors are used in the model. This helps account for errors that follow seasonal patterns.
For Beginners: This is similar to the regular Q value, but for seasonal prediction errors. With the default value of 1, the model considers how wrong its prediction was for the same period last season. For example, if last December's sales forecast was too high, the model might adjust this December's forecast downward. This helps the model learn from seasonal forecasting mistakes.
Tolerance
Gets or sets the convergence tolerance for the optimization algorithm.
public double Tolerance { get; set; }Property Value
- double
The convergence tolerance, defaulting to 0.00001 (1e-5).
Remarks
The algorithm stops when the improvement between iterations is smaller than this tolerance value, indicating that the model has converged to a solution.
For Beginners: This determines when the model decides it's "good enough" and stops trying to improve. The default value of 0.00001 means the model will stop when additional training produces very tiny improvements (less than one hundred-thousandth). Think of it like painting a wall - at some point, adding another coat of paint makes such a small difference that it's not worth the effort. This parameter defines that "barely noticeable difference" threshold.