A new insight into prediction modeling system

1
A new insight into prediction modeling system

• Sang C. Suh
• Sam I. Saffer
• Dan Li (Presenter)
• Jingmiao
• Department of Computer Science
• Texas AM University-Commerce
• Commerce, Texas 75429-3011, USA

IDPT 2003, Austin, TX, December 5
2
Overview

• Objective of our paper
• Intelligent Forecasting Model Selection System
• Our focus is Time Series Analysis
• No one knows the FUTURE
• All forecasting (prediction) methods or
  techniques are used to help make decisions
• All models are wrong, but some are useful.
• --- George Box (1994)

3
Background

• Forecast Trend Season Cycle Random
• Trend a long-term upward or downward change in
  the time series
• Seasonal periodic increases / decreases that
  occur within a year
• Cycle periodic increases / decreases that occur
  over more than a one-year period
• Random (Irregular, Stochastic ) changes in the
  time series not attributable to the other three
  components

4
Introduction on different methods

• Time Series Techniques
• - Smoothing
• - Fourier Series Analysis
• - ARIMA
• Other statistics/quantitative/DM methods
• - Regressions (MRA-OLS)
• - Decision trees
• - ANNs

5
Introduction (cont)

• Smoothing Methods
• SMA
• SES
• Browns a, b ? trend, level
• Holts a,ß? trend, level
• Winters a,ß, I ? trend, level, seasonality

6
Comparisons Analysis

• There is no single best forecasting model.
• Each model may be best fitted into specific
  situation such as horizon length, automation of
  development, pattern recognition ability, number
  of observation required, etc.
• Prediction Selection is condition-dependent

7
(No Transcript)
8
(No Transcript)
9
Prediction Modeling System
10
Prediction Modeling System (cont)

• Pattern Identification Module (PIM)
• - ACF, PACF, Mean, SD, RUNS,etc
• Pattern Comparison Module (PCM)
• - ANNs based on pattern classifications
• Model Selection Module (MSM)
• - Rule-based Expert System
• Error Check Module (ECM)
• - RSE, MSE, RMSE, MAPE, etc.
• Model Comparison Module (MCM)
• - only dealing with 2 or more candidates

11
Pattern Identification Module (PIM)

• Identify Trend and/or Seasonality

12
Pattern Identification Module (PIM)

• This module is to use statistic tools to analyze
  data sequence first for patterns classification.
  We built auto-correlation function (ACF) and
  partial auto-correlation function (PACF) to
  facilitate the objective and use t-value to test
  the coefficients.
• ACF measures the direction and strength of the
  statistical relationship between ordered pairs of
  observations on two random variables. It measures
  how closely the matched pairs are related to each
  other.
• PACF measures the correlation between ordered
  pairs separated by various time spans (k1, 2, 3
  ) with the effects of intervening observations
  accounted for.

13
Auto-Correlation Function (ACF)
14
Partial Auto-Correlation Function (PACF)
15
Pattern Comparison Module (PCM)

• This module is to do the pattern comparison by
  using back-propagation neural network. We use
  neural network instead of human experts to learn
  and train neurons (statistic results from PIM),
  eliminate outliers and output the trend of the
  statistic data sequence, which are added pattern
  information. This neural network based module can
  intelligently generate pattern information such
  as no season, with trend, stationary, etc.

16
Model Pattern Classifications
17
Model Selection Module (MSM)

• This module is to select the available right
  model (s) from the pattern information generated
  from PCM based on the rule-based pattern table.
  For example, based on no season, with trend,
  stationary time series data sequence, Holts
  Two-Parameter Trend Model may be a right choice.
  This model is a pure knowledge based expert
  system.

18
MSM Demo
19
Error Check Module (ECM)

• RSE (Relative Standard Error)
• MSE (Mean Square Error)
• RMSE (Root Mean Square Error)
• MAPE (Mean Absolute Percentage Error)

20
Model Comparison Module (MCM)

• This is a complementary module
• It is used only when there are two or more model
  candidates selected from MSM. Then it has to
  repeat both models to compare the forecasting
  accuracy by residual analysis also.

21
Conclusions and future work

• There was not one method that was best for all
  series or all forecast horizons.
• Each forecasting method has its own criterion,
  assumptions, constraints.
• There are around 30-40 forecasting methods in
  current world.
• And lots of researches in hybrid methods.
• Our Prediction Modeling System can be further
  extended and optimized.

22
THANK YOU ?

• Sang C. Suh, Ph.D.
• Sam I. Saffer, Ph.D.
• Dan Li
• Jingmiao Gao
• Department of Computer Science
• Texas AM University-Commerce
• Commerce, Texas 75429-3011, USA
• (All reference list can be available from our
  paper)
• Made in Texas, USA, 12/2/2003