Upload
phungdien
View
214
Download
2
Embed Size (px)
Citation preview
© 2017 Sven F. Crone. All rights reserved.
Beyond Predictive Analytics?Delivering value on in business forecasting
for Virgin Atlantic Cargo
Dr. Sven F. CroneIIF Workshop on Predictive Analytics & Forecasting, Munich, 14.09.2017
© 2017 Sven F. Crone. All rights reserved.
• The Hype?• From Forecasting to Analytics?• Gartner’s Hype Cycle over time• The Gap in Forecasting
• Analytics at Virgin Cargo!• Descriptive Analytics• Predictive Analytics• (towards) Prescriptive Analytics
• Feedback?
Analytics at Virgin Cargo
Agenda
© 2017 Sven F. Crone. All rights reserved.
Inventory
Optimization
Demand Sensing
Demand Shaping
Demand Shifting
Outside-in
Focused
Proactive
Process
Lean Consensus
Forecasting
FVA
Market Supplier
Demand
Driven
Inside-out
Focused
Reactive
Process
Supply Sensing
Supply Shaping
Sales Marketing & Operations Planning
Market
Driven
Selling through the channel (pull) Selling into the channel (push)
Supply
Driven
15-30%
5-7%
Lean
Manufacturing
Lean
Forecasting
FVA
ShipmentsSales OrdersPOS
The “Market-Driven” Supply Chain Journey
© Charlie Chase @ SAS! talk to him to get his presentations
© 2017 Sven F. Crone. All rights reserved.
Is most value-added by
PrescriptiveAnalytics aka
Demand Shaping?
33%
13%
3%
[Gartner Magic Quadrant for BI platforms, 2012]
100% BI Implementations
Analytics Value Escalator
Which level of„predictive“?
univaritate ETS vs. Causal Neural
Networks?Data Exploration
Demand Forecasting
Demand Shaping
???
???%
?%
??%
Hype-Cycle for Emerging Technologies
New Technologies make bold promises
how to discern the hype
from what's commercially viable?Should you make an early move?
Take a moderate approach?
Wait for further maturation?
Are we all on
the plateau
of productivity?
In 2015 Gartner dropped Big Data,
Analytics, and Data Science!
© 2017 Sven F. Crone. All rights reserved.
• The Hype?• From Forecasting to Analytics?• Gartner’s Hype Cycle over time• The Gap in Forecasting
• Analytics at Virgin Cargo!• Descriptive Analytics• Predictive Analytics• (towards) Prescriptive Analytics
• Feedback?
Analytics at Virgin Cargo
Agenda
Destinations: 31
Fleet size: 39 (+ 23 orders)
Revenue: £2.93 billion (2014)
Passengers: 5,5 Millionen (2012)
The Virgin Atlantic Story
• Inaugural flight from London to Newark took place on 22nd June 1984.
• Grown in the 27 years since to become the UK’s 2nd largest long haul airline
• Set up an in-house department to handle its cargo product shortly afterwards.
© 2017 Sven F. Crone. All rights reserved.
Virgin CargoAirbus 330/340 and Boeing
747/777/787 series aircraft.
3,200 LD3 units
© 2017 Sven F. Crone. All rights reserved.
Original Foreasting @ Virgin?
ARIMA black-box( VB-plug-in,developed by another University)
© 2017 Sven F. Crone. All rights reserved.
Flight Rollup-Hierarchical Structure
Region
Sub-region
Area
Sector
Does not identify inbound/outbound nodes which are bottlenecks
© 2017 Sven F. Crone. All rights reserved.
Seasonality at different frequencies?
Total Network Quarterly Seasonality
Total Network Weekly Seasonality
Total Network Yearly Seasonality
© 2017 Sven F. Crone. All rights reserved.
• The Hype?• From Forecasting to Analytics?• Gartner’s Hype Cycle over time• The Gap in Forecasting
• Analytics at Virgin Cargo!• Descriptive Analytics• Predictive Analytics• (towards) Prescriptive Analytics
• Feedback?
Analytics at Virgin Cargo
Agenda
© 2017 Sven F. Crone. All rights reserved.
Predictive Analytics
Normal Forecasting with Exponential Smoothing
Smoothes out everything, including Christmas, Easter …
© 2017 Sven F. Crone. All rights reserved.
Conventional Forecasting
Seasonality can handle some effects: Christmas etc.
Seasonality cannot handle Easter & Chinese New Year moves!
© 2017 Sven F. Crone. All rights reserved.
Calendar EventsBrazil Independence from PortugalAlso known as: Dia da Independência (Portugese)A national holiday observed in Brazil on September 7 of every year.The date celebrates Brazil's Declaration of Independence from Portugal on September 7, 1822.Chinese New Year (Day)Encodes only the 2nd week of CNY (if event spans two weeks) together with a small lead effect in the previous weekChristmas 26Gregorian New YearNew Year on January 1st.Korea Traditional New YearAlso known as: Gujeong (Korean), Seol-nal (Korean)Traditional Korean New Year festival celebrated on the first day of the lunar calendarThailand Traditional New YearAlso known as: Wan Songkran (Thai)The Songkran festival is celebrated in Thailand as the traditional New Year's Day from 13 to 15 April.USA ThanksgivingCelebrated on the 4th Thursday in November. Traditional start of the holiday season.Traditionally celebrates the giving of thanks for the autumn harvest.
Detected Events can be modelled as binary dummy variables
Extra time series of 0 and 1 where an event occurs
© 2017 Sven F. Crone. All rights reserved.
detected events in past & future captures holidays & season
NARX(p) Neural Network with Calendar Events
© 2017 Sven F. Crone. All rights reserved.
External (causal?) Drivers
Source: IATA, Seabury
Note: FTK stands for Freight Tonne Kilometres, the industry standard demand metric. The chart shows how this metric
has moved over the 2000-2012 period.
© 2017 Sven F. Crone. All rights reserved.
Next steps? Prescriptive Analytics!
Expand Models with more variables?• Internal factors, eg. Capacity (different hullspace)
• External factors, eg. GDP growth, industry leading indicators, Global PMI, OECD Confidence index etc.
© 2017 Sven F. Crone. All rights reserved.
Analytics Value escalator
Fo
recast A
ccu
racy
Naive
Averages
ETSTriple ETS Neural Nets
simple k-NN
Diminishing returnsin accuracy on
growing complexityof algorithms
© 2017 Sven F. Crone. All rights reserved.
• The Hype?• From Forecasting to Analytics?• Gartner’s Hype Cycle over time• The Gap in Forecasting
• Analytics at Virgin Cargo!• Descriptive Analytics• Predictive Analytics• (towards) Prescriptive Analytics
• Feedback?
Analytics at Virgin Cargo
Agenda
© 2017 Sven F. Crone. All rights reserved.
A Multinomial logit optimization model
75%
75%
75%
Fractional Logit
1a Total TV Model
1b TV Market Share Model
2 Optimization Model
1 Aggregate Demand
Objective Function:
q: Unit Sales
t: Week
pl: Price Level
i: Brandπ: Market share
p: Price
α: Attraction
Parameter
β: Price
Parameter
Simple Regression
Total price level = weighted by market shares
c: Cost
g: Profit
Partially following González-Benito et al. (2010)
no consideration of the
Heteroskedasticity (yet!)
Komplex approach …
How to price Weekly Chargable Weight per Route? (for maximum profit)?
© 2017 Sven F. Crone. All rights reserved.
Analytics Value Escalator
PrescriptiveAnalytics
DescriptiveAnalytics
PredictiveAnalytics
Is most value-add
generated byPrescriptiveAnalytics???
Limited
capability
to execute perceived
value
actual
value
potential
value
training!
Extend your
Tool Box!Descriptive Tasks
using Data Visualisation
Predictive Models including
Artificial Neural Networks,
Support Vector Regression
C4.5, k-Means, SVM, Apriori, EM,
PageRank, AdaBoost, kNN,
Naive Bayes, and CART
Prescriptive Tasks in
Inventory Managment,
Revenue Management etc.
Artificial Intelligencemeets Forecasting.
Mulitlayer Perceptrons, RecurrentNeural Nets, Extremem Learning Machines Decision Trees, k-nearest NeighboursSupport Vector Machines, MARS/EARTH + statistical benchmarks in
NOW HIRING:
•PhD-Thesis
•Master-Thesis
•Werkstudenten
•Summer Internships
•Data sponsoring for PhDs
•Data Scientists!
•PhD students (with Lancs)
Take aways
• The hype is real– … but it‘s just a hype you can use!
• The gap is substantial. – Companies are slow adopters to algorithms– (not all) Software companies are innovators
• The potential is substantial– Calendar effects can easily be foreasted
• Important on an agregate level of regions / subregions• Improves accuracy significantly for important days
– Requires more complex algorithms• Introduces causal models of regression or neural networks• Could also incoporate prices, number of flights, aircraft sizes …• Available in standard software (at approx. 5000.-)
– Potential for best practices in logistics forecasts– Try new algorithms!
• Neural Networks• Support Vector Regression• Decision Trees• K-Nearest Neighbours• …
© 2017 Sven F. Crone. All rights reserved.
© 2017 Sven F. Crone. ALL RIGHTS RESERVEDUnless otherwise indicated, all material in this presentation is copyrighted by Sven F. Crone. No part of these pages, either text or images may be used for any purpose other than personal use, unless explicit authorisation is given by the author.
Reproduction, modification, storage in a retrieval system or retransmission, in any form or by any means – electronic or otherwise, for reasons other than personal use, is strictly prohibited without prior permission.
1. The content of the pages of this presentation is for your general information and use only. It is subject to change without notice.
2. Neither we nor any third parties provide any warranty or guarantee as to the accuracy, timeliness, performance, completeness or suitability of the information and materials found or offered in this presentation for any particular purpose. You acknowledge that such information and materials may contain inaccuracies or errors and we expressly exclude liability for any such inaccuracies or errors to the fullest extent permitted by law.
3. Your use of any information or materials in this presentation is entirely at your own risk, for which we shall not be liable. It shall be your own responsibility to ensure that any products, services or information identified in this presentation meet your specific requirements.
4. This presentation contains material which is owned by or licensed to us. This material includes, but is not limited to, the design, layout, look, appearance and graphics. Reproduction is prohibited other than in accordance with the copyright notice, which forms part of these terms and conditions.
5. All trade marks reproduced in this website which are not the property of, or licensed to, the operator are acknowledged in the presentation.
6. Unauthorised use of this presentation may give rise to a claim for damages and/or be a criminal offence.
7. From time to time this presentation may also include links to other presentations or websites. These links are provided for your convenience to provide further information. They do not signify that we endorse the other presentations or website(s). We have no responsibility for the content of the linked presentations or website(s).
8. Your use of this presentation and any dispute arising out of such use of the presentation is subject to the laws of Hamburg, Germany.
9. No images rendered can be used for mechanical, electronic or printed reproduction without prior written permission by thee author. Images generated are for previewing purposes only.