Table of Contents

Enum TimeSeriesModelType

Namespace
AiDotNet.Enums
Assembly
AiDotNet.dll

Represents different types of time series forecasting models used for analyzing and predicting sequential data over time.

public enum TimeSeriesModelType

Fields

ARIMA = 0

Auto-Regressive Integrated Moving Average model - a standard statistical method for time series forecasting that combines autoregression, differencing, and moving average components.

For Beginners: ARIMA is like a Swiss Army knife for time series forecasting - it's versatile and widely used.

ARIMA looks at three aspects of your data:

  • AR (Auto-Regressive): How much the current value depends on previous values
  • I (Integrated): How many times we need to subtract consecutive values to make the data stable
  • MA (Moving Average): How much random "noise" from previous time points affects the current value

ARIMA is typically written as ARIMA(p,d,q) where:

  • p = number of lag observations (AR component)
  • d = degree of differencing (I component)
  • q = size of the moving average window (MA component)

When to use it: When your data doesn't have strong seasonal patterns and you need a reliable, well-established forecasting method.

Example: Forecasting monthly product sales that show a general trend but no clear seasonal pattern.

ARIMAX = 12

ARIMA model with additional explanatory variables (exogenous variables) that can influence the forecast.

For Beginners: ARIMAX is like ARIMA but allows you to include external factors that might affect your forecast.

Standard ARIMA only looks at past values of the thing you're trying to predict. ARIMAX lets you add other relevant variables. For example, when forecasting ice cream sales, you could include temperature as an external factor.

The "X" in ARIMAX stands for "exogenous variables" - factors outside your main time series that influence it.

When to use it: When you know external factors influence what you're trying to predict.

Example: Forecasting energy consumption while accounting for temperature, day of week, and holidays.

ARMA = 2

Auto-Regressive Moving Average model - combines autoregressive and moving average components without the differencing (integration) step.

For Beginners: ARMA is a simpler version of ARIMA used when your data is already stable.

ARMA combines:

  • AR (Auto-Regressive): How past values influence the current value
  • MA (Moving Average): How past random fluctuations influence the current value

The key difference from ARIMA is that ARMA assumes your data is already "stationary" (meaning its statistical properties don't change over time).

When to use it: When your data doesn't show strong trends that change over time.

Example: Analyzing stable temperature fluctuations around a consistent average.

AutoRegressive = 3

Auto-Regressive model - predicts future values based solely on past values of the same variable.

For Beginners: AutoRegressive models predict the future based only on the past values of the data itself.

Think of AutoRegressive like predicting tomorrow's temperature based only on today's and yesterday's temperatures. The model assumes that recent past values have the strongest influence on the next value.

AR(p) means we're using the previous p values to make our prediction.

When to use it: When you believe recent past values strongly influence future values.

Example: Predicting stock prices where recent price movements tend to influence short-term future prices.

BayesianStructuralTimeSeriesModel = 18

A flexible Bayesian approach to time series modeling that incorporates prior knowledge and uncertainty.

For Beginners: This model uses Bayesian statistics to incorporate uncertainty and prior knowledge into forecasts.

Bayesian models are different because they:

  • Start with "prior beliefs" about what might happen
  • Update these beliefs as new data comes in
  • Express results as probability distributions rather than single-point forecasts

The "structural" part means the model breaks down your time series into components like trend, seasonality, and cycle components.

When to use it: When you have prior knowledge about your data that you want to incorporate, or when understanding the uncertainty in your forecast is important.

Example: Economic forecasting where you want to incorporate expert knowledge and clearly communicate the range of possible outcomes and their probabilities.

Custom = 24

Represents a custom or user-defined time series model not covered by the standard types.

For Beginners: This option allows you to implement your own specialized time series model.

The Custom option gives you flexibility to:

  • Implement models not included in the standard list
  • Create hybrid approaches combining multiple techniques
  • Develop domain-specific models tailored to your particular data

When to use it: When standard models don't meet your specific needs or when you want to implement a novel approach to time series analysis.

Example: Implementing a specialized forecasting model designed specifically for your industry or unique data characteristics.

DoubleExponentialSmoothing = 7

An extension of simple exponential smoothing that can handle data with a trend component. Also known as Holt's method.

For Beginners: Double Exponential Smoothing adds the ability to handle data that shows an upward or downward trend.

While Simple Exponential Smoothing works for data that fluctuates around a stable average, Double Exponential Smoothing (also called Holt's method) can handle data that's consistently increasing or decreasing over time.

It uses two smoothing parameters:

  • Alpha: Controls how much recent levels affect the forecast
  • Beta: Controls how much recent trends affect the forecast

When to use it: When your data shows a consistent upward or downward trend.

Example: Forecasting a company's growing monthly revenue that shows a steady upward trend.

DynamicRegressionWithARIMAErrors = 11

A model that combines regression with ARIMA modeling of the error terms to account for both external factors and time dependencies.

For Beginners: This model combines traditional regression with time series techniques.

Imagine you're forecasting ice cream sales. You know temperature affects sales (higher temperature = higher sales), but there are also time-based patterns. This model lets you:

  1. Use regression to account for the temperature effect
  2. Use ARIMA to model the remaining patterns in your data

When to use it: When external factors influence your time series and you want to account for both these factors and time-based patterns.

Example: Forecasting product sales while accounting for price changes, marketing spend, and seasonal patterns.

ExponentialSmoothing = 5

A general class of forecasting methods that give more weight to recent observations and less weight to older observations.

For Beginners: Exponential Smoothing is like having a weighted average where newer data points matter more than older ones.

Imagine you're trying to predict tomorrow's temperature. You might care more about today's temperature than what happened a week ago. Exponential smoothing does exactly this - it gives more importance to recent observations and less to older ones.

This is a general category that includes several specific methods (simple, double, and triple).

When to use it: When recent observations are more relevant to your prediction than older ones.

Example: Forecasting customer demand where recent purchasing patterns are more relevant than historical ones.

GARCH = 13

Generalized Autoregressive Conditional Heteroskedasticity model - specialized for forecasting volatility in time series.

For Beginners: GARCH models are designed specifically for predicting how much a value will fluctuate or vary over time.

While most time series models focus on predicting the actual values, GARCH focuses on predicting the volatility (how much the values jump around). This is especially useful in finance where understanding risk (volatility) is often as important as predicting prices.

GARCH models recognize that periods of high volatility tend to cluster together - if today is highly volatile, tomorrow is likely to be volatile too.

When to use it: When you care about forecasting the variability or uncertainty in your data, not just the values themselves.

Example: Forecasting stock market volatility to assess investment risk.

InterventionAnalysis = 21

Analyzes how specific events or interventions affect a time series and quantifies their impact.

For Beginners: Intervention Analysis helps measure how specific events changed your time series.

This approach focuses on quantifying the impact of known events or "interventions" on your data. For example:

  • How did a marketing campaign affect sales?
  • What was the impact of a policy change on crime rates?
  • How did a website redesign affect user engagement?

The analysis typically compares what actually happened after the intervention with what would have happened without it (the "counterfactual").

When to use it: When you want to measure the impact of specific events or actions on your time series.

Example: Analyzing how a price change affected product demand by comparing actual sales after the change with the predicted sales if no change had occurred.

MA = 4

Moving Average model - predicts future values based on past forecast errors rather than past values.

For Beginners: MA models focus on the errors or surprises in previous predictions.

Instead of using past values directly, MA uses past prediction errors (the difference between what we predicted and what actually happened). It's like saying "I was off by X yesterday, so I should adjust my prediction today."

MA(q) means we're using the previous q prediction errors to make our forecast.

When to use it: When random shocks or events have lingering effects on your data.

Example: Call center volume forecasting where unexpected spikes (like after a product issue) affect call volumes for several days.

NeuralNetworkARIMA = 17

A hybrid model that combines neural networks with traditional ARIMA models to leverage the strengths of both approaches.

For Beginners: This hybrid model combines traditional statistical methods with modern machine learning.

Think of this as a "best of both worlds" approach:

  • ARIMA provides a solid statistical foundation that works well for linear patterns
  • Neural Networks add the ability to capture complex, non-linear patterns

The hybrid approach typically works by:

  1. Using ARIMA to model the linear components of your data
  2. Using neural networks to capture the remaining non-linear patterns

When to use it: When your data contains both simple linear patterns and complex non-linear relationships that a single model type might miss.

Example: Forecasting energy consumption where there are both predictable patterns (like daily cycles) and complex relationships with multiple factors (weather, events, etc.).

ProphetModel = 16

A forecasting model developed by Facebook that handles multiple seasonality patterns and is robust to missing data and outliers.

For Beginners: Prophet is a user-friendly forecasting tool designed by Facebook to be easy to use while still being powerful.

Prophet works like an advanced version of decomposition methods - it breaks down your time series into:

  • Trend: The overall direction (increasing or decreasing)
  • Seasonality: Regular patterns at different time scales (daily, weekly, yearly)
  • Holiday effects: Irregular but predictable events

What makes Prophet special:

  • It's robust to missing data and outliers
  • It can automatically detect changepoints (where trends shift)
  • It handles multiple seasonal patterns well
  • It allows you to incorporate domain knowledge easily

When to use it: When you need reliable forecasts without extensive time series expertise, especially for data with strong seasonal patterns or irregular events.

Example: Forecasting business metrics like website traffic or product demand that have weekly patterns, yearly seasonality, and holiday effects.

SARIMA = 1

Seasonal Auto-Regressive Integrated Moving Average model - extends ARIMA to handle data with seasonal patterns.

For Beginners: SARIMA is like ARIMA but with added capabilities to handle predictable seasonal patterns.

If your data shows regular patterns that repeat (like higher ice cream sales every summer), SARIMA can capture both the overall trend and these seasonal fluctuations.

SARIMA adds seasonal components to the standard ARIMA model and is typically written as SARIMA(p,d,q)(P,D,Q)m where:

  • (p,d,q) are the non-seasonal parameters (same as ARIMA)
  • (P,D,Q) are the seasonal parameters
  • m is the number of time periods in each season (e.g., 12 for monthly data with yearly seasonality)

When to use it: When your data shows clear seasonal patterns that repeat at regular intervals.

Example: Retail sales data that shows holiday shopping spikes every December.

STLDecomposition = 20

Seasonal and Trend decomposition using Loess - breaks down time series into trend, seasonal, and remainder components.

For Beginners: STL Decomposition is like separating a smoothie back into its original ingredients.

This method breaks your time series into three components:

  • Trend: The long-term progression (increasing or decreasing)
  • Seasonality: Regular patterns that repeat at fixed intervals
  • Remainder: What's left after removing trend and seasonality (irregular fluctuations)

The "Loess" part refers to a statistical method used to estimate the trend component through local regression.

When to use it: When you want to understand the different components driving your time series or remove seasonality before further analysis.

Example: Breaking down retail sales data to separate the overall growth trend from seasonal holiday patterns and random fluctuations.

SimpleExponentialSmoothing = 6

The most basic form of exponential smoothing that handles data with no clear trend or seasonality.

For Beginners: Simple Exponential Smoothing is the most basic version that works well for stable data without trends.

This method calculates a weighted average of all past observations, with weights decreasing exponentially as observations get older. It's controlled by a single parameter (alpha) that determines how quickly the influence of past observations decays.

When to use it: When your data fluctuates around a stable average with no clear upward/downward trend or seasonal patterns.

Example: Forecasting stable inventory levels for products with consistent demand.

SpectralAnalysis = 19

Analyzes time series data by decomposing it into different frequency components to identify cyclical patterns.

For Beginners: Spectral Analysis is like breaking down a song into individual notes to understand its composition.

This approach transforms time series data from the time domain to the frequency domain. Instead of asking "what happens next?", it asks "what cycles or rhythms exist in this data?"

Imagine your data as a complex musical chord - spectral analysis breaks it down into the individual notes (frequencies) that make up that chord.

When to use it: When you want to identify hidden cycles or periodic patterns in your data, especially when multiple cycles might be overlapping.

Example: Analyzing sunspot activity to identify various solar cycles, or analyzing economic data to identify business cycles of different lengths.

StateSpace = 9

A flexible framework for time series modeling that represents a system's behavior using state variables.

For Beginners: State Space models track the "hidden state" of a system that we can't directly observe.

Imagine you're trying to track a person's location based only on their cell phone signal strength. You can't directly see where they are (the hidden state), but you can make educated guesses based on the signal strength you observe.

State Space models work similarly for time series - they try to uncover the underlying state of the system that's generating your observable data.

When to use it: When you believe there are underlying factors driving your data that aren't directly observable.

Example: Tracking the true economic health of a country based on various economic indicators.

TBATS = 10

A flexible time series model that handles complex seasonal patterns using trigonometric components. TBATS stands for Trigonometric, Box-Cox transform, ARMA errors, Trend, and Seasonal components.

For Beginners: TBATS is a specialized model for handling complex or multiple seasonal patterns.

While models like SARIMA work well for simple seasonal patterns, TBATS can handle more complex situations like:

  • Multiple seasonal patterns (e.g., daily, weekly, and yearly patterns all at once)
  • Changing seasonal patterns
  • Non-integer seasonality (e.g., 365.25 days in a year)

TBATS uses mathematical techniques (trigonometric functions) to represent these complex patterns.

When to use it: When your data has multiple or complex seasonal patterns.

Example: Hourly electricity demand data that shows daily patterns, weekly patterns, and yearly seasonal patterns all at once.

TransferFunctionModel = 22

Models how one time series affects another with potential time delays between cause and effect.

For Beginners: Transfer Function Models help understand how one variable affects another over time.

These models are designed to capture how changes in an input variable (X) affect an output variable (Y) over time, including:

  • How strong the effect is
  • How long it takes for the effect to appear (delay)
  • How long the effect lasts

Unlike simple correlation, transfer functions can model complex relationships where effects are spread out over time.

When to use it: When you want to model how one time series influences another, especially when there are time delays between cause and effect.

Example: Modeling how advertising expenditure affects sales over time, where spending today might influence sales for several weeks or months.

TripleExponentialSmoothing = 8

An extension of double exponential smoothing that can handle data with both trend and seasonal components. Also known as Holt-Winters' method.

For Beginners: Triple Exponential Smoothing handles data with both trends and seasonal patterns.

This method (also called Holt-Winters) is the most comprehensive of the exponential smoothing family. It can model:

  • The overall level of the data
  • Upward or downward trends
  • Seasonal patterns that repeat at regular intervals

It uses three smoothing parameters:

  • Alpha: Controls how much recent levels affect the forecast
  • Beta: Controls how much recent trends affect the forecast
  • Gamma: Controls how much recent seasonal patterns affect the forecast

When to use it: When your data shows both a trend and regular seasonal patterns.

Example: Forecasting ice cream sales that show both an overall increasing trend and higher sales every summer.

UnobservedComponentsModel = 23

Models time series by representing them as combinations of unobserved components like trend, cycle, and seasonality.

For Beginners: This model breaks down your time series into hidden components that can't be directly observed.

Unobserved Components Models (UCM) assume your time series is made up of several underlying components that you can't directly measure, such as:

  • Trend: The long-term direction
  • Cycle: Medium-term fluctuations
  • Seasonality: Regular patterns that repeat
  • Irregular: Random fluctuations

The model uses statistical techniques to estimate these hidden components from your observable data.

When to use it: When you want a flexible framework for decomposing your time series into meaningful components for analysis or forecasting.

Example: Analyzing economic indicators by separating long-term growth trends from business cycles and seasonal patterns.

VAR = 14

Vector Autoregression model - extends autoregressive models to multiple related time series that influence each other.

For Beginners: VAR models handle multiple related time series that affect each other.

While models like ARIMA work with a single time series, VAR handles multiple related series simultaneously. For example, prices of related products might influence each other - if beef prices rise, chicken demand (and then prices) might also increase as consumers switch.

VAR captures these interactions and lets each variable be influenced by its own past values AND the past values of other variables in the system.

When to use it: When you have multiple time series that influence each other.

Example: Analyzing how changes in interest rates, inflation, and unemployment affect each other over time.

VARMA = 15

Vector Autoregression Moving-Average model - combines VAR and moving average components for multiple related time series.

For Beginners: VARMA extends the VAR model by adding moving average components.

VARMA combines two approaches:

  • Vector Autoregression (VAR): How past values of multiple related variables affect current values
  • Moving Average (MA): How past random shocks or surprises affect current values

This gives VARMA more flexibility to capture complex relationships between multiple time series.

When to use it: When you have multiple related time series with complex interactions that aren't fully captured by simpler models.

Example: Analyzing interactions between economic indicators like GDP, unemployment, and inflation where both past values and unexpected events influence each other.

Remarks

For Beginners: Time series models help us understand patterns in data that change over time and make predictions about future values.

Think of time series data as any measurement collected regularly over time - like daily temperature readings, monthly sales figures, or hourly website traffic. These models help us answer questions like:

  • "What will our sales be next month?"
  • "How many visitors will our website get tomorrow?"
  • "What will the temperature be next week?"

Different models are designed to capture different patterns in time data:

  • Some are good at finding seasonal patterns (like holiday shopping spikes)
  • Others excel at detecting long-term trends (like gradual population growth)
  • Some can handle sudden changes or outliers (like a viral social media post)

The right model depends on your specific data and what patterns you expect to find in it.

Edit this page