Time series forecasting algorithm parameters
Each description includes the algorithm(s) to which it applies.
Adjust Outliers
Used in: Automatic, Automatic Plus, Arima, ArimaX, Auto Arima, Exponential Smoothing, Intermittent, Model Average Neural Network, Naive Forecasting, Quantile Random Forest, Simple Moving Average, Stochastic Gradient Boosting, SVM Linear, Tournament
Description: Adjust outliers during execution to improve fit quality
Type: Boolean
Range: NA
Valid Values: TRUE, FALSE
Default Value: FALSE for Time Series Forecasting, TRUE for Life Cycle Modeling
Bagging
Used in: Model Average Neural Network
Description: If TRUE, each repeat of a neural network trains on a sample of the training data.
Type: SingleValue (Logical)
Range: N/A
Valid Values: TRUE, FALSE
Default Value: FALSE
Recommended Value: FALSE
Cost
Used in: SVM Linear
Description: The cost regularization parameter; very large value attempts to fit every point and can lead to overfitting, very small value can lead to underfitting
Type: SingleValue (Numeric)
Range: N/A
Valid Values: C > 0
Default Value: 1.0
Recommended Value: 1
Demand Correlation Threshold
Used in: Automatic, Automatic Plus, Arima, ArimaX, Auto Arima, Exponential Smoothing, Intermittent, Model Average Neural Network, Naive Forecasting, Quantile Random Forest, Simple Moving Average, Stochastic Gradient Boosting, SVM Linear, Tournament
Description: Threshold for absolute correlation with demand between demand and a causal
Type: Double
Range: 0-1
Valid Values: NA
Default Value: 0.15
Include Causal Lags
Used in: Automatic, Automatic Plus, Arima, ArimaX, Auto Arima, Exponential Smoothing, Intermittent, Model Average Neural Network, Naive Forecasting, Quantile Random Forest, Simple Moving Average, Stochastic Gradient Boosting, SVM Linear, Tournament
Description: Include lag effects following causals
Type: Boolean
Range: NA
Valid Values: TRUE, FALSE
Default Values: TRUE
Is Centered
Used in: Simple Moving Average
Description: If FALSE, consider the most recent ‘Moving Average Type’ number of observations. If TRUE, consider the ‘Moving Average Type’ number of observations from the center value of the time series (by time) when taking the average.
Type: SingleValue (Logical)
Range: N/A
Valid Values: TRUE, FALSE
Default Value: FALSE
Method Name
Used in: Intermittent
Description: Croston-type intermittent method by which to perform demand forecasting
Type: SingleValue (Character)
Range: N/A
Valid Values: auto, croston, sba, sbj
Default Value: auto
Recommended Value: auto
Minimum Terminal Node Size
Used in: Stochastic Gradient Boosting
Description: Minimum number of observations in the terminal nodes of the trees
Type: SingleValue (Int32)
Range: N/A
Valid Values: 5-25
Default Value: 10
Recommended Value: 10
Model Names
Used in: Tournament
Description: A list of models to be considered for finding out the best model based on Out-Of-Sample RMSE
Type: List (Character)
Range: N/A
Valid Values: A list of models to be considered for determining the best model based on Out-Of-Sample RMSE
Default Values: SVMLinear, SimpleMovingAverage, ExponentialSmoothing
Moving Average Type
Used in: Simple Moving Average
Description: The number of past observations to be considered
Type: SingleValue (Int)
Range: N/A
Valid Values: N/A
Default Value: Auto
Recommended Value: A number between 4-8
Multi Collinearity Threshold
Used in: Automatic, Automatic Plus, Arima, ArimaX, Auto Arima, Exponential Smoothing, Intermittent, Model Average Neural Network, Naive Forecasting, Quantile Random Forest, Simple Moving Average, Stochastic Gradient Boosting, SVM Linear, Tournament
Description: Threshold for the absolute correlation coeffecient between two causals
Type: Double
Range: 0-1
Valid Values: NA
Default Value: 0.95
Non-seasonal Auto Regressive Lag
Used in: Arima
Description: Moving backward in the time series history, the number of lags to be considered, where each lag represents demand value from a previous period
Type: SingleValue (Int32)
Range: 0-10
Valid Values: N/A
Default Value: 0
Non-seasonal Difference
Used in: Arima
Description: The number of times to perform differencing between consecutive values of a non-stationary time series to create a new stationary time series in which mean, variance, and autocorrelation do not change over time
Type: SingleValue (Int32)
Range: 0-10
Valid Values: N/A
Default Value: 0
Non-seasonal Moving Average Lag
Used in: Arima
Description: The number of differences (error values) to be considered between actual data points and corresponding fitted values in the demand history
Type: SingleValue (Int32)
Range: 0-10
Valid Values: N/A
Default Value: 1
Number of Boosting Operations
Used in: Stochastic Gradient Boosting
Description: The total number of trees to fit. Equivalent to the number of iterations.
Type: SingleValue (Int32)
Range: NA
Valid Values: 1-5000
Default Value: Auto
Recommended Value: 50 or 100
Number of Hidden Units
Used in: Model Average Neural Network
Description: Number of units hidden in the layer
Type: SingleValue (Int32)
Range: N/A
Valid Values: N/A
Default Value: Auto
Recommended Value: 3 or 4
Number of Randomly Selected Causal Variables
Used in: Quantile Random Forest
Description: Number of variables randomly sampled at each node of each tree
Type: SingleValue (Int32)
Range: N/A
Valid Value: Less than the number of causal variables
Default Value: Auto
Recommended Value: 2 or 3
Number of Splits
Used in: Stochastic Gradient Boosting
Description: The number of splits performed on a tree
Type: SingleValue (Int32)
Range: NA
Valid Values: 1-10
Default Value: Auto
Recommended Value: 2 or 3
Seasonal Auto Regressive Lag
Used in: Arima
Description: Moving backward in the time series history, the number of seasonal demand lags to be considered, where each lag represents demand value from a previous season
Type: SingleValue (Int32)
Range: 0-10
Valid Values: N/A
Default Value: 0
Seasonal Difference
Used in: Arima
Description: The number of times to perform differencing across seasons between consecutive values of a non-stationary time series to create a new stationary time series in which mean, variance, and autocorrelation do not change over time
Type: SingleValue (Int32)
Range: 0-10
Valid Values: N/A
Default Value: 0
Seasonal Frequency
Used in: Arima, Auto Arima, Exponential Smoothing
Description: Seasonal frequency in data, identifying the number of data points after which a pattern in the data is repeating
Type: SingleValue (Int32)
Range: 1-52
Valid Values: 1-52
Default Values: NA
Recommended Value for Auto Arima and Exponential Smoothing: 7 for daily data, 26 or 52 for weekly data, 12 for monthly data, 4 for quarterly data, 1 for yearly data
Seasonal Moving Average Lag
Used in: Arima
Description: The number of differences (error values) to be considered between actual data points and corresponding fitted values across seasons in the demand history
Type: SingleValue (Int32)
Range: 0-10
Valid Values: N/A
Default Value: 0
Shrinkage
Used in: Stochastic Gradient Boosting
Description: The learning rate or step-size reduction applied to each tree.
Type: SingleValue (Numeric)
Range: N/A
Valid Values: 0.001 to 0.6
Default Value: 0.1
Recommended Value: 0.1
Weight Decay
Used in: Model Average Neural Network
Description: Parameter for weight decay to help avoid overfitting
Type: SingleValue (Numeric)
Range: N/A
Valid Values: N/A
Default Value: Auto
Recommended Value: 0.01 or 0.05
Last modified: Thursday December 19, 2024