Class InterventionAnalysisOptions<T, TInput, TOutput>
Configuration options for Intervention Analysis, which is a time series modeling technique used to assess the impact of specific events or interventions on a time series.
public class InterventionAnalysisOptions<T, TInput, TOutput> : TimeSeriesRegressionOptions<T>Type Parameters
TThe numeric data type used for calculations (typically double or float).
TInputTOutput
- Inheritance
-
InterventionAnalysisOptions<T, TInput, TOutput>
- Inherited Members
Remarks
Intervention Analysis extends time series regression by explicitly modeling the effects of known interventions or events (like policy changes, marketing campaigns, or natural disasters) on a time series. It combines ARIMA (AutoRegressive Integrated Moving Average) modeling with additional components that represent these interventions, allowing you to quantify their impact while accounting for the underlying time series patterns.
For Beginners: Intervention Analysis helps you measure how specific events affected your time series data. For example, if you're tracking daily sales and launched a major marketing campaign on a specific date, intervention analysis can help you determine how much that campaign actually boosted your sales while accounting for other factors like seasonal patterns or existing trends.
Think of it like trying to measure the effect of a new medication on a patient's health while accounting for their normal day-to-day fluctuations in health metrics. Just looking at the raw numbers before and after might be misleading - intervention analysis gives you a more accurate picture of the true impact.
This class inherits from TimeSeriesRegressionOptions, so all the general time series regression settings are also available. The additional settings specific to intervention analysis let you configure how the model handles both the time series patterns and the intervention effects.
Properties
AROrder
Gets or sets the order of the AutoRegressive (AR) component in the ARIMA model.
public int AROrder { get; set; }Property Value
- int
The AR order, defaulting to 1.
Remarks
The AR order specifies how many previous time points directly influence the current value in the time series model. Higher values allow the model to capture more complex patterns of autocorrelation but increase the risk of overfitting and computational complexity.
For Beginners: This setting determines how many previous time periods your model should look at to predict the current value. With the default value of 1, the model considers only the immediately previous time period when making predictions.
Think of it like predicting tomorrow's weather based on today's weather (AR order = 1) versus predicting it based on today's, yesterday's, and the day before yesterday's weather (AR order = 3). A higher AR order can capture more complex patterns but might also pick up on random fluctuations that don't actually help with prediction.
For example, with daily sales data:
- AR order = 1: Today's sales depend on yesterday's sales
- AR order = 7: Today's sales depend on sales from each of the previous 7 days
Start with the default value of 1 and increase it if you believe there are longer-term dependencies in your data. For many business time series, values between 1 and 7 are common, often corresponding to dependencies on the previous day, week, or month.
Interventions
Gets or sets the list of interventions to be analyzed in the time series.
public List<InterventionInfo> Interventions { get; set; }Property Value
- List<InterventionInfo>
A list of intervention information objects, defaulting to an empty list.
Remarks
Each intervention is defined by its timing (when it occurred) and type (how it affected the time series). Common intervention types include pulse (temporary effect), step (permanent level change), and ramp (gradual change). The model will estimate the magnitude and significance of each intervention's effect.
For Beginners: This is where you specify the events or changes you want to analyze. Each intervention needs information about when it happened and what type of effect you expect it to have.
For example, if you're analyzing website traffic data and launched a new marketing campaign on June 1, 2023, you would add an intervention with that date. You'd also specify whether you expect:
- A temporary spike in traffic (pulse intervention)
- A permanent increase in the baseline traffic level (step intervention)
- A gradual increase that builds over time (ramp intervention)
You can add multiple interventions to analyze several events at once. For instance, you might want to analyze both a marketing campaign and a website redesign that happened on different dates.
The model will then estimate how much each intervention actually affected your time series, accounting for other factors like seasonal patterns and existing trends.
MAOrder
Gets or sets the order of the Moving Average (MA) component in the ARIMA model.
public int MAOrder { get; set; }Property Value
- int
The MA order, defaulting to 1.
Remarks
The MA order specifies how many previous error terms (differences between predicted and actual values) influence the current value in the time series model. Higher values allow the model to account for more complex patterns in the error terms but increase the risk of overfitting and computational complexity.
For Beginners: This setting determines how many previous prediction errors your model should consider when making the current prediction. With the default value of 1, the model adjusts its current prediction based on how wrong its previous prediction was.
While the AR component looks at previous actual values, the MA component looks at previous prediction errors. This helps the model correct for systematic errors it might be making.
Think of it like a weather forecaster who not only looks at previous weather (AR) but also adjusts based on how wrong their recent forecasts were (MA). If they consistently underestimated temperatures last week, they might adjust this week's predictions upward.
For example, with daily sales data:
- MA order = 1: Today's prediction is adjusted based on yesterday's prediction error
- MA order = 3: Today's prediction is adjusted based on prediction errors from the past 3 days
The default value of 1 works well for many applications. Higher values might help if there are complex patterns in how your model's errors behave over time.
Optimizer
Gets or sets the optimizer used to find the best parameters for the intervention analysis model.
public IOptimizer<T, TInput, TOutput>? Optimizer { get; set; }Property Value
- IOptimizer<T, TInput, TOutput>
The optimizer instance, defaulting to null (which will use a default optimizer).
Remarks
The optimizer is responsible for finding the parameter values that minimize the error between the model's predictions and the actual time series values. Different optimizers may have different performance characteristics depending on the specific time series and interventions being analyzed.
For Beginners: This setting lets you choose the algorithm that will be used to find the best possible model parameters. If you leave it as null (the default), the system will choose an appropriate optimizer for you.
Think of the optimizer as the "learner" that figures out exactly how much each intervention affected your time series and how strong the time series patterns are. Different optimizers use different strategies to find these values, similar to how different people might use different strategies to find the lowest price for a product.
Most beginners should leave this as null and let the system choose an appropriate optimizer. Advanced users might want to specify a particular optimizer if they have specific requirements regarding speed, precision, or if they're dealing with unusual time series patterns.