Variable Selection Based on Stepwise Regression
Variable Selection Based on the SAS Enterprise Miner Variable Selection Node
Variable Selection Based on the SAS Enterprise Miner Partial Least Squares Node
Variable Selection Based on Decision Trees
Variable Selection Based on Genetic Programming
Comparison of Data Mining Variable Selection Results
7.4 Time Series Approach
7.5 Summary
Chapter 8 Model Building: ARMA Models
Introduction
8.1 ARMA Models
8.1.1 AR Models: Concepts and Application
8.1.2 Moving Average Models: Concepts and Application
8.1.3 Auto Regressive Moving Average (ARMA) Models
Appendix 1: Useful Technical Details
Appendix 2: The “I” in ARIMA
Chapter 9 Model Building: ARIMAX or Dynamic Regression Modes
Introduction
9.1 ARIMAX Concepts
9.2 ARIMAX Applications
Appendix: Prewhitening and Other Topics Associated with Interval-Valued Input Variables
Chapter 10 Model Building: Further Modeling Topics
Introduction
10.1 Creating Time Series Data and Data Hierarchies Using Accumulation and Aggregation Methods
Introduction
Creating Time Series Data Using Accumulation Methods
Creating Data Hierarchies Using Aggregation Methods
10.2 Statistical Forecast Reconciliation
10.3 Intermittent Demand
10.4 High-Frequency Data and Mixed-Frequency Forecasting
High-Frequency Data
Mixed-Interval Forecasting
10.5 Holdout Samples and Forecast Model Selection in Time Series
Introduction
10.6 Planning Versus Forecasting and Manual Overrides
10.7 Scenario-Based Forecasting
10.8 New Product Forecasting
Chapter 11 Model Building: Alternative Modeling Approaches
11.1 Nonlinear Forecasting Models
11.1.1 Nonlinear Modeling Features
11.1.2 Forecasting Models Based on Neural Networks
11.1.3 Forecasting Models Based on Support Vector Machines
11.1.4 Forecasting Models Based on Evolutionary Computation
11.2 More Modeling Alternatives
11.2.1 Multivariate Models
11.2.2 Unobserved Component Models (UCM)
Chapter 12 An Example of Data Mining for Forecasting
12.1 The Business Problem
12.2 The Charter
12.3 The Mind Map
12.4 Data Sources
12.5 Data Prep
12.6 Exploratory Analysis and Data Preprocessing
12.7 X Variable Imputation
12.8 Variable Reduction and Selection
12.9 Modeling
12.10 Summary
Preface
It is utterly impossible that a mathematical formula should make the future known to us, and those who think it can would once have believed in witchcraft.
Jacob Bernoulli, in Ars Conjectadi, 1713
Curiosity about “what will happen next” is part of human nature, and thus the first attempts at forecasting are found rooted in history. In the ancient and medieval times, prophets like the Oracle of Delphi or Nostradamus had the status of demigods. The situation is significantly different in the 21st century, though, when predicting the future is not divine magic anymore but a necessity in contemporary business. Thousands of professionals are building forecasts in almost all areas of human activity. Since the global recession of 2008–2009, it has been much more widely understood that reliable forecasting is necessary.
The increased demand for forecasting triggered the development of new methods in addition to the “classical” time series statistical approaches, such as exponential smoothing and the Box-Jenkins AutoRegressive Integrated Moving-Average (ARIMA) models. One fruitful direction of development is that of nonlinear time series modeling, based on various computational intelligence methods, such as neural networks, support vector machines, and genetic programming. Other developments, of special importance to industrial applications, are the efforts for improving the time series forecasts by selecting the best potential drivers
