SAS^® Visual Forecasting Features

Large-scale time series analysis & forecasting in a distributed environment

• Automatically generates large quantities of statistically based forecasts in a distributed, in-memory environment.
• Scripting language enables distributed, in-memory time series analysis. 
• Shuffles the data so that each time series is copied into the memory of a single computing node.
• Executes each time series on one thread of a node, and each node executes the compiled script for each of its assigned time series.
• Is optimized for the machine on which it is running, so users don’t have to rewrite code for different machines.

Neural network/machine learning modeling strategy nodes 

• Includes a panel series neural network framework for generating features and training a neural network.
• Provides a multistage (neural network/regression + time series) framework for creating a forecasting methodology that combines signals from different types of models.
• Addresses problems that have both time series characteristics and nonlinear relationships between dependent and independent variables using stacked model (neural network + time series) forecasting.

Flexible override facility

• Enables customized forecast adjustments that aren't limited by the structure of the forecasting hierarchy.
• Lets you select filters based on attributes, such as location, brand, category, size, color, sentiment, quality, etc.
• Lets you define override specifications by filter and time period(s) for all time series contained within a filter.
• Includes faceted search filters.
• Allows disaggregation of override using optimization model.
• Enables batch execution and incremental data updates.

Support for open source languages

• Includes External Language Package (EXTLANG), which supports open source code – Python and R.
• Lets you call analytical actions from Python, R, Java, JavaScript and Lua.
  

Time series analysis

• Autocorrelation analysis.
• Cross-correlation analysis.
• Seasonal decomposition and adjustment analysis.
• Count series analysis.
• Diagnostic tests for seasonality, stationarity, intermittency and tentative ARMA order selection.
  

Time frequency analysis

• Windowing functions.
• Fourier analysis for real and complex time series.
• Short-time Fourier analysis.
• Discrete Hilbert transform.
• Pseudo Wigner-Ville distribution.

Time series modeling

• ARIMA models (dynamic regression and transfer functions). 
• Exponential smoothing models.
• Unobserved component models. 
• State-space models. 
• Intermittent demand models with Croston’s method.
  

Automatic time series modeling

• Automatic time series model generation. 
• Automatic input variable and event selection. 
• Automatic model selection. 
• Automatic parameter optimization.
• Automatic forecasting.

Singular spectrum analysis (SSA)

• Univariate SSA decomposition and forecasting.
• Multivariate SSA. 
• Automatic SSA.

Subspace tracking (SST)

• Perform advanced monitoring (signal analysis) techniques for multiple time series.

Time interval evaluation

• Evaluate a variable in an input table for suitability as a time ID variable.
• Assess how well a time interval specification fits date/datetime values or observation numbers used to index a time series. 
• Can either be specified explicitly as input to PROC TSMODEL or inferred by the procedure based on values of the time ID variable. 
  

Hierarchical reconciliation

• Models and forecasts each series in the hierarchy individually.
• Reconciles forecasts at multiple levels of the hierarchy.

Derived attributes

• Create predefined sets of derived attributes, including:
  • Time series attributes (min, max, mean, missing, etc.).
  • Forecasting attributes (model properties, statistics of fit).
  • Demand Classification attributes.
  • Volume/Volatility attributes.
• Attributes are available for use in faceted search in the viewers.

External segmentation & demand classification

• Provides support for 1,000 segments.
• Can utilize predefined segments in time series.
• Includes a prebuilt segmentation template based on Demand Classification.

Time series & forecast viewers

• Provides a Time Series Viewer with prebuilt set of time series attributes.
• Provides a Forecast Viewer with prebuilt set of forecasting attributes.
• Includes envelope plots for viewing multiple series.
• Lets you use faceted filters on descriptive statistics, model properties, statistics of fit.

Time Series Dimension Reduction (TDR) package

• Enables dimension reduction of transactional time series data in preparation for time series mining.
  • Lets you then apply traditional data mining techniques (such as clustering, classification, decision trees and others).

Distributed, accessible & cloud-ready

• Runs on SAS^® Viya^®, a scalable and distributed in-memory engine of the SAS Platform.
• Distributes analysis and data tasks across multiple computing nodes.
• Provides fast, concurrent, multiuser access to data in memory.
• Includes fault tolerance for high availability.
• Lets you add the power of SAS Analytics to other applications using SAS Viya REST APIs.