Universität Hamburg Institut für Wirtschaftsinformatik Prof. Dr. D.B. Preßmar Final Results of the NN3 Neural Network Forecasting Competition Sven F. Crone,

Slides:



Advertisements
Similar presentations
Prof. Carolina Ruiz Department of Computer Science Worcester Polytechnic Institute INTRODUCTION TO KNOWLEDGE DISCOVERY IN DATABASES AND DATA MINING.
Advertisements

Part II – TIME SERIES ANALYSIS C3 Exponential Smoothing Methods © Angel A. Juan & Carles Serrat - UPC 2007/2008.
1 Why the damped trend works Everette S. Gardner, Jr. Eddie McKenzie.
Bootstrapping judgemental adjustments to improve forecasting accuracy - judgemental bootstraps vs error bootstraps Robert Fildes Centre for Forecasting,
Tim Smyth and Jamie Shutler Assessment of analysis and forecast skill Assessment using satellite data.
Department of Industrial Management Engineering 1.Introduction ○Usability evaluation primarily summative ○Informal intuitive evaluations by designers even.
Database Implementation of a Model-Free Classifier Konstantinos Morfonios ADBIS 2007 University of Athens.
SBSE Course 3. EA applications to SE Analysis Design Implementation Testing Reference: Evolutionary Computing in Search-Based Software Engineering Leo.
LRS Progress Report and Action Plan Update to the Profiling Working Group March 30, 2006.
Dr. Yukun Bao School of Management, HUST Business Forecasting: Experiments and Case Studies.
Experimental Evaluation in Computer Science: A Quantitative Study Paul Lukowicz, Ernst A. Heinz, Lutz Prechelt and Walter F. Tichy Journal of Systems and.
Task1: Predicting Citations Claudia Perlich, Foster Provost & Sofus Macskassy New York University KDD Cup, August 2003.
Chapter 12 - Forecasting Forecasting is important in the business decision-making process in which a current choice or decision has future implications:
Engineering Data Analysis & Modeling Practical Solutions to Practical Problems Dr. James McNames Biomedical Signal Processing Laboratory Electrical & Computer.
Analysis of Classification-based Error Functions Mike Rimer Dr. Tony Martinez BYU Computer Science Dept. 18 March 2006.
ForecastingOMS 335 Welcome to Forecasting Summer Semester 2002 Introduction.
Scientific method - 1 Scientific method is a body of techniques for investigating phenomena and acquiring new knowledge, as well as for correcting and.
Experimental Evaluation in Computer Science: A Quantitative Study Paul Lukowicz, Ernst A. Heinz, Lutz Prechelt and Walter F. Tichy Journal of Systems and.
OECD Short-Term Economic Statistics Working PartyJune Analysis of revisions for short-term economic statistics Richard McKenzie OECD OECD Short.
Application of seasonal climate forecasts to predict regional scale crop yields in South Africa Trevor Lumsden and Roland Schulze School of Bioresources.
1 Prediction of Software Reliability Using Neural Network and Fuzzy Logic Professor David Rine Seminar Notes.
Attention Deficit Hyperactivity Disorder (ADHD) Student Classification Using Genetic Algorithm and Artificial Neural Network S. Yenaeng 1, S. Saelee 2.
Evaluation of software engineering. Software engineering research : Research in SE aims to achieve two main goals: 1) To increase the knowledge about.
The impact of mobility on productivity and career path (WP7) Aldo Geuna Cornelia Meissner Paolo Cecchelli University of Torino Fondazione Rosselli.
1 Pathways to a World-Class Competitive Intelligence Function John E. Prescott, Ph.D Alessandro Comai, Ph.D candidate (ESADE Business.
Algoval: Evaluation Server Past, Present and Future Simon Lucas Computer Science Dept Essex University 25 January, 2002.
Xiao Liu, Jinjun Chen, Ke Liu, Yun Yang CS3: Centre for Complex Software Systems and Services Swinburne University of Technology, Melbourne, Australia.
Simon van Norden HEC Montréal and CIRANO Marc Wildi Institute of Data Analysis and Process Design, Winterthur.
Cost drivers, cost behaviour and cost estimation
Frontiers in the Convergence of Bioscience and Information Technologies 2007 Seyed Koosha Golmohammadi, Lukasz Kurgan, Brendan Crowley, and Marek Reformat.
Exploiting Context Analysis for Combining Multiple Entity Resolution Systems -Ramu Bandaru Zhaoqi Chen Dmitri V.kalashnikov Sharad Mehrotra.
Science in Business Data Mining? Background: support managerial decision making Background: support managerial decision making Is there a science to data.
Estimating Component Availability by Dempster-Shafer Belief Networks Estimating Component Availability by Dempster-Shafer Belief Networks Lan Guo Lane.
1 STAT 5814 Statistical Data Mining. 2 Use of SAS Data Mining.
Qi Guo Emory University Ryen White, Susan Dumais, Jue Wang, Blake Anderson Microsoft Presented by Tetsuya Sakai, Microsoft Research.
Omer Levy Yoav Goldberg Ido Dagan Bar-Ilan University Israel
WERST – Methodology Group
Iterative similarity based adaptation technique for Cross Domain text classification Under: Prof. Amitabha Mukherjee By: Narendra Roy Roll no: Group:
Hybrid Load Forecasting Method With Analysis of Temperature Sensitivities Authors: Kyung-Bin Song, Seong-Kwan Ha, Jung-Wook Park, Dong-Jin Kweon, Kyu-Ho.
AGU Fall Meeting 2008 Multi-scale assimilation of remotely sensed snow observations for hydrologic estimation Kostas Andreadis, and Dennis Lettenmaier.
Of An Expert System.  Introduction  What is AI?  Intelligent in Human & Machine? What is Expert System? How are Expert System used? Elements of ES.
Artificial Intelligence for Data Mining in the Context of Enterprise Systems Thesis Presentation by Real Carbonneau.
Stock market forecasting using LASSO Linear Regression model
Development of an Ensemble Gridded Hydrometeorological Forcing Dataset over the Contiguous United States Andrew J. Newman 1, Martyn P. Clark 1, Jason Craig.
Downscaling Global Climate Model Forecasts by Using Neural Networks Mark Bailey, Becca Latto, Dr. Nabin Malakar, Dr. Barry Gross, Pedro Placido The City.
G W. Yan 1 Multi-Model Fusion for Robust Time-Series Forecasting Weizhong Yan Industrial Artificial Intelligence Lab GE Global Research Center.
1 Exponential smoothing in the telecommunications data Everette S. Gardner, Jr.
LOAD FORECASTING. - ELECTRICAL LOAD FORECASTING IS THE ESTIMATION FOR FUTURE LOAD BY AN INDUSTRY OR UTILITY COMPANY - IT HAS MANY APPLICATIONS INCLUDING.
RELATION EXTRACTION, SYMBOLIC SEMANTICS, DISTRIBUTIONAL SEMANTICS Heng Ji Oct13, 2015 Acknowledgement: distributional semantics slides from.
PCWG-Share-01 Current Status PCWG Meeting Hamburg 10 th March 2016 Peter Stuart (RES) and Andy Clifton (NREL), on behalf of the Power Curve Working Group.
AAAI Spring Symposium : 23 March Brenna D. Argall : The Robotics Institute Learning Robot Motion Control from Demonstration and Human Advice Brenna.
Prediction intervals for ensemble time series forecasts
Deep Learning for Dual-Energy X-Ray
The Big Data for Official Statistics Competition
An Artificial Intelligence Approach to Precision Oncology
Table 1. Advantages and Disadvantages of Traditional DM/ML Methods
A Time Series Representation Framework Based on Learned Patterns
AI in Cyber-security: Examples of Algorithms & Techniques
Application of satellite-based rainfall and medium range meteorological forecast in real-time flood forecasting in the Upper Mahanadi River basin Trushnamayee.
Software Engineering Experimentation
Data, Economics and Computational Agricultural Science
What barriers and facilitators influence the implementation of new high-risk medicine services in Scottish community pharmacies? Ms Natalie Weir1, Dr Rosemary.
The Open World of Micro-Videos
Toward a Reliable Evaluation of Mixed-Initiative Systems
Chap. 7 Regularization for Deep Learning (7.8~7.12 )
Mohamed Dirir, Norma Sinclair, and Erin Strauts
Supervisor: Yury Nikulin Key research questions:
Fundamentals of Neural Networks Dr. Satinder Bal Gupta
Q: How can we make people believe our predictions/projections?
Implementation of a small-scale desktop grid computing infrastructure in a commercial domain    
Presentation transcript:

Universität Hamburg Institut für Wirtschaftsinformatik Prof. Dr. D.B. Preßmar Final Results of the NN3 Neural Network Forecasting Competition Sven F. Crone, Konstantinos Nikolopoulos and Michele Hibon

 Can NN modelling be automated for business forecasting?  Evaluate progress in NN modelling since M3  Disseminate Explicit knowledge on “best practices” 2005 SAS & IIF Grant RATIONAL OBJECTIVES RESULTS DISCUSSION FURTHER RESEARCH

2005 SAS & IIF Grant RATIONAL

 Only 1 evaluation of NN within Forecasting Competitions  Distinct fields of research and participation NN: breakthrough or passing fad? Reid 1969 Santa Fe 1991 BUSINESS FORECASTING COMPETITIONS NN COMPETITIONS Suykens 1998 Reid 1972 Newbold & Granger 1974 Makridakis & Hibon 1979 M-Competition 1982 M2-Competition 1988 M3-Competition 2000 H-Competition, Hibon 2006 EUNITE 2001 ANNEXG 2001 BI Cup 2003 CATS 2005 ISF ISF06 ANNEX 2006 WCCI 2006 Only 1 NN entry Balkin & Ord

 Most NN competitions = classification ( EUNITE’02, WCCI06 etc.)  Limited evidence on Regression evaluations  Visit CI Time Series Competitions Time SeriesData FormatLengthSubmis. SANTA FE 1991 Gershenfeld & Weigend 2 univariate 4 multivariate UV: Laser, UV: Artificial, Sleep, Exchange rate, Astrophysics, Music 1000, 34000, , , 27704, Black Box 1998 Suykens & Vandewalle 1 univariatePhysics 2000 (1000) 17 EUNITE multivariateElectrical Load (31) 56 ANNEXG 2001 Dawson et al. 1 multivariateHydrology 1460 points Hydrology 12 BI Cup 2003 Weber 1 multivariateSugar sales 365 days (14) 10+ CATS 2005, IEEE Lendasse, 1 univariate in 5 parts Artificial 4905pointas (95 points, 5*19) 25 ISF2005 Crone 2 univariateAirline, M3-Competition144, 8516 ANNEXG / ISF2006 Dawson et al., Crone 3 multivariateHydrology 1460 points Hydrology 12 WCCI 2006 Predictive Uncertainty, Gawley 1 univariate 3 multivariate UV: Synthetic, Precipitation, Temperature, SO2 380, 10000, 10000,

 Conduct competition on industry data  Evaluate different NN methodologies  Can NN forecasting be AUTOMATED on many time series? Reasons? Modelling Decisions Gap between forecasting & NN domains  NN evaluations on different data types  No positive evidence on M-type data Short time series Noisy time series Discouraging research findings  NN cannot forecast seasonal time series  No valid & reliable methodology to model NN  No automation of NN modelling possible

 Can NN modelling be automated for business forecasting?  Evaluate progress in NN modelling since M3  Disseminate Explicit knowledge on “best practices” 2005 SAS & IIF Grant OBJECTIVES a)What is the performance (accuracy, robustness & resources) of NN in comparison to established forecasting methods? b)What are the current “best practice” methodologies utilised by researchers to model NN for time series forecasting

 Multiple Hypothesis Testing similar to M3-competition Competition Design Multiple empirical Time Series  Complete set of 111 time series  Reduced set of 11 time series  Representative structures  monthly industry data long & short time series Seasonal and non-seasonal series  Scaled observations for anonymity  No domain knowledge  18 steps ahead forecasts Simulated ex ante (out of sample) evaluation Multiple error measures & computational time Testing of conditions under which NN perform well/bad NN3 COMPETITION

Competition Design 46 Submissions for the reduced dataset 9 benchmarks 22 submissions for the complete dataset 8 benchmarks Submissions

2005 SAS & IIF Grant Automated AI/CI approaches can very well do the job! (batch forecasting) Balkin’s and Ord approach was not very ‘bad’ after all.. Performance was verified across many metrics (including MASE), parametric + non-parametric Performance was verified with multiple hypothesis: long/short, seasonal/non seasonal, difficult/easy So… WHAT do we know NOW that we did not knew before NN3?

2005 SAS & IIF Grant Time Series Benchmarks are very hard to beat! Forecast Pro, Theta model and Marc Wildi’s Stat benchmark outperform overall all CI/AI approaches For the ‘harder’ part of the NN3 dataset – 25 short+non-seasonal series – CI approaches managed to outperform all other approaches!! Full automation seems to be possible in large scale forecasting tasks + Side results… New Stat benchmarks that perform outstandingly Improvement of established forecasting engines in the last 10 years So… WHAT do we know NOW that we did not knew before NN3?

+

+ +

+

2005 SAS & IIF Grant Computational times…. Leaders of the field (Academia + Commercial) Time series features that would necessitate the use of AI/CI approaches Replication in a competition of the M3 volume (NN5…111, tourism competition…1000+) Best practices? Full automation?? and… WHAT we still DO NOT …

? Sven, Kostas & Michele