9th IFAC Conference on Manufacturing Modelling, Management and

9th
9th IFAC
IFAC Conference
Control Conference on
on Manufacturing
Manufacturing Modelling,
Modelling, Management
Management and
and
9th IFAC
Control
9th IFAC Conference
Conference on
on Manufacturing
Manufacturing Modelling,
Available
Modelling, Management
online and
at www.sciencedirect.com
Management and
Control
Berlin,
9th IFACGermany, August
Conference on 28-30, 2019
Manufacturing Modelling, Management and
Control
Berlin,
Control Germany, August 28-30, 2019
Berlin,
Control Germany, August 28-30, 2019
Berlin, Germany, August 28-30, 2019
Berlin, Germany, August 28-30, 2019
Berlin, Germany, August 28-30, 2019 ScienceDirect
IFAC PapersOnLine 52-13 (2019) 737–742
Machine
Machine Learning
Learning in
in Predicting
Predicting Demand
Demand for
for
Machine
Machine
Fast-Moving Learning
Learning
Consumer in
in Predicting
Predicting
Goods: An Demand
Demand
Exploratory for
for
Research
Fast-Moving Consumer
Machine
Fast-Moving Goods:
Learning in An Exploratory
Predicting Demand Research
for
Fast-Moving Consumer Goods: An Exploratory Research
Consumer Goods: An Exploratory Research
Fast-Moving Consumer Goods: An Exploratory Research
Elcio Tarallo* Getúlio K. Akabane** Camilo I. Shimabukuro***
Elcio
Elcio Tarallo*
Tarallo* Getúlio
Getúlio K.
K. Akabane**
Akabane** Camilo
Camilo I. I. Shimabukuro***
I.*****
Shimabukuro***
Elcio
Elcio Tarallo*
Tarallo* Jose MelloK.
Getúlio
Getúlio ****
K. Douglas Amancio
Akabane**
Akabane** Camilo
Camilo I. Shimabukuro***
Shimabukuro***
Elcio Tarallo* Jose
Jose Mello
Mello
Getúlio ****
****
K. Douglas
Douglas
Akabane** Amancio
Amancio
Camilo I. *****
*****
Shimabukuro***
Jose
Jose Mello
Mello ****
**** Douglas
Douglas Amancio
Amancio *****
*****
*Centro
Jose Mello Paula
**** Souza,
Douglas SãoAmancio
Paulo, Brazil *****
*Centro
*Centro Paula Souza,
Paula Souza, SãoSão Paulo,
Paulo, Brazil
Brazil
(Tel: +55 11 99153-0039;
*Centro
*Centro Paula
Paula e-mail:São
Souza,
Souza, [email protected])
São Paulo,
Paulo, Brazil
Brazil
(Tel:
(Tel: +55
+55Souza,11
11 99153-0039;
99153-0039;
*Centro Paula e-mail:
e-mail:
Souza, [email protected])
[email protected])
São Paulo,[email protected])
Brazil
**Centro Paula
(Tel:
(Tel: +55
+55 11
11 São
99153-0039;
99153-0039; Paulo, Brazil
e-mail:
e-mail: (e-mail:
[email protected])
[email protected])
**Centro
**Centro Paula
Paula
(Tel: +55 Souza,
Souza,
11 São
São
99153-0039; Paulo,
Paulo, Brazil
Brazil
e-mail: (e-mail:
(e-mail: [email protected])
[email protected])
[email protected])
*** Centro
**Centro
**Centro Paula
Paula
Paula Souza,
Souza,
Souza, São São
SãoPaulo,
Paulo,
Paulo, Brazil (email:
Brazil
Brazil (e-mail:
(e-mail: [email protected])
[email protected])
[email protected])
***
*** Centro
Centro
**Centro Paula
Paula
Paula Souza,
Souza,
Souza, São
São SãoPaulo,
Paulo,
Paulo, Brazil
Brazil (email:
(email:
Brazil (e-mail: [email protected])
[email protected])
[email protected])
***
******* Centro
Centro
Centro Paula
PaulaPaula Souza,
Souza,
Souza, São
São São
Paulo,
Paulo,Paulo, Brazil
Brazil
Brazil (e-mail:
(email:
(email: [email protected])
[email protected])
[email protected])
*******
**** Centro
Centro
Centro PaulaPaula
Paula Souza,
Souza,
Souza, São São
São
Paulo,Paulo,
Paulo, Brazil
Brazil
Brazil (e-mail:
(e-mail:
(email: [email protected])
[email protected])
[email protected])
*********
**** Centro
Centro
Centro Paula
Paula
Paula Souza,
Souza,
Souza, São
São
São Paulo,
Paulo,
Paulo, Brazil
Brazil
Brazil (e-mail:
(e-mail:
(e-mail: [email protected])
[email protected])
[email protected])
*****
********* Centro
Centro
Centro Paula
Paula
Paula Souza,
Souza,
Souza, São
São
São Paulo,
Paulo,
Paulo, Brazil
Brazil
Brazil (e-mail:
(e-mail:
(e-mail: [email protected])
[email protected])
[email protected])
*****
***** Centro
Centro PaulaPaula Souza,
Souza, São São Paulo,
Paulo, Brazil
Brazil (e-mail:
(e-mail: [email protected])
[email protected])
***** Centro Paula Souza, São Paulo, Brazil (e-mail: [email protected])
Abstract: More accurate prediction of the demand for fast-moving consumer goods is a competitive
Abstract: More
Abstract: More accurate
accurate prediction
prediction of of the
the demand
demand for for fast-moving
fast-moving consumer consumer goods goods is is aa competitive
competitive
factor
Abstract: for manufacturers
More accurateand
More accurate retailers,of
prediction
prediction ofespecially
the demand
the demandin the fashion,
forfashion,
for fast-moving
fast-moving technology consumer
consumer and fresh
goodsfood
goods food sectors.
is aasectors.
is competitive
competitiveThis
factor
Abstract:
factor for
for manufacturers
manufacturers
More accurate and
and retailers,
retailers,
prediction ofespecially
especially
the demandin
in the
the
forfashion,
fast-moving technology
technology consumer and
and fresh
fresh
goods food
is a sectors.
competitiveThis
This
exploratory
Abstract:
factor
factor for research
for manufacturers
manufacturers presents
and the
and retailers, benefits
retailers, of
especially
especially Machine
in
in the Learning
fashion, in
fashion, technology
the Learning sales
technology forecasting
and
and fresh
fresh forfor
food
food short shelf-life
sectors.
sectors. This
This
exploratory
exploratory research
research presents
presents the
the benefits of Machine in sales forecasting short shelf-life
factor
and
exploratory
exploratory
for manufacturers
highly-perishable
research
research products,
presents
presents the
theas benefits
and retailers, it surpasses
benefits
benefits
of
especially
of
of
Machine
Machine
Machine
the Learning
theinaccuracy
fashion,
Learning
Learning
in
leveltechnology
in
in
sales
of traditional
sales
sales
forecasting
forecasting
forecasting
fresh for
andstatistical food
for
for
short
short
short
shelf-life
sectors.
techniques This
and,
shelf-life
shelf-life
and
and highly-perishable
highly-perishable
exploratory research products,
products,
presents theas
as it
it surpasses
surpasses
benefits of the
the accuracy
accuracy
Machine level
level
Learning of
of
in traditional
traditional
sales statistical
statistical
forecasting for techniques
techniques
short and,
and,
shelf-life
as
and
and a result, improves
highly-perishable
highly-perishable inventory
products,
products, balancing
as
as it
it surpasses
surpassesthroughout
the
the the
accuracy
accuracy chain,
level
level reducing
of
of stockout
traditional
traditional rates
statistical
statistical at points
techniques
techniques of sale,
and,
and,
as
as
and a
a result,
result, improves
improves
highly-perishable inventory
inventory
products, balancing
balancing
as it surpassesthroughout
throughout
the the
the
accuracy chain,
chain,
level reducing
reducing
of stockout
stockout
traditional rates
rates
statistical at
at points
points
techniques of
of sale,
sale,
and,
improving
as
as a
a result,
result, availability
improves
improves to consumers
inventory
inventory and
balancing
balancing increasing
throughout
throughout profitability.
the
the chain,
chain, Copyright
reducing
reducing © 2019
stockout
stockout IFAC
rates
rates at
at points
points of
of sale,
sale,
improving
improving
as a result, availability
availability
improves to
to consumers
consumers
inventory and
and
balancing increasing
increasing
throughout profitability.
profitability.
the chain, Copyright
Copyright
reducing ©
© 2019
2019
stockout IFAC
IFAC
rates at points of sale,
improving
improving
© 2019, IFAC availability
availability to
to consumers
(International consumers
Federation and
and increasing
Automatic profitability.
increasing
of prediction, profitability.
Control) HostingCopyright
Copyright ©
© 2019
by Elsevier 2019Ltd.IFAC
IFAC
All rights reserved.fast-
Keywords:
improving Machine
availability Learning,
to consumers demand and increasing sales forecast,
profitability. short
Copyright shelf-life,
© 2019 short
IFAC life-cycle,
Keywords:
Keywords: Machine
Machine Learning,
Learning, demand
demand prediction,
prediction, sales
sales forecast,
forecast, short
short shelf-life,
shelf-life, short
short life-cycle,
life-cycle, fast-
fast-
moving
Keywords: consumer
Machine
Machine goods.
Learning,
Learning, demand
demand prediction,
prediction, sales
sales forecast,
forecast, short
short shelf-life,
shelf-life, short life-cycle,
short life-cycle, fast-
fast-
moving
Keywords:
moving
Keywords: consumer
consumer
Machine goods.
goods.
Learning, demand prediction, sales forecast, short shelf-life, short life-cycle, fast-
moving
moving consumer
consumer goods. goods.
moving consumer goods.
1. INTRODUCTION In the retail food industry in general, the main cause of
1. INTRODUCTION In the
In the retail
retail food food industry
industry in in general,
general, the the main
main cause cause of of
1. wasted products and stockouts is the inaccuracy of sales
1. INTRODUCTION
1. INTRODUCTION
INTRODUCTION
In
In the
the retail
wasted
wasted retail
products
products
food
foodand
and
industry
industry
stockouts
stockouts
in
in general,
general,
is
is the
the
the
the main
main cause
inaccuracy
inaccuracy cause
of
of
of
of
sales
sales
Demand forecasting1. is a basic component of production In
INTRODUCTION the retail
forecasting
wasted
wasted products
products
foodand
leading and
industry
to incorrect
stockouts
stockouts
in orders
general,
is
is the
the
the main
(Arunraj
inaccuracy
inaccuracy andcause
Ahrens,
of
of
of
sales
sales
Demand
Demand forecasting
forecasting is
is a
a basic
basic component
component of
of production
production forecasting
forecasting
wasted leading
leading
products andto
to incorrect
incorrect
stockouts orders
orders
is the (Arunraj
(Arunraj
inaccuracy and
and Ahrens,
Ahrens,
of sales
planning
Demand and supply chain management, impacting 2015).
forecastingMore specifically
leading to in the
incorrect fresh
orders food industry,
(Arunraj and including
Ahrens,
planning forecasting
Demand forecasting
and supply supply is
is aa chain
basic
basic component
chain component
management, of
of production
impacting forecasting
production 2015). More
2015). Moreleading to incorrect
specifically
specifically in the
in orders
the fresh
fresh food(Arunraj
industry, andincluding
Ahrens,
including
planning
Demand
competitiveness
planning and
forecasting
and andis aprofitability,
supply chain management,
basic component
management,providing impacting
of productioncritical forecasting
impacting the
2015).
2015). refrigerated
More
More
leading to incorrect
ones,
specifically
specifically such
in
in the
the the food
orders
asfresh
fresh dairy,
food
food
industry,
(Arunrajfruitand
industry,
industry, andAhrens,
juice
including
including
planning
competitiveness
competitivenessand supply
and
and chain
profitability,
profitability,management,providing
providing impacting
critical
critical the
the refrigerated
refrigerated ones,
ones, such
such as
as the
the dairy,
dairy, fruit
fruit and
and juice
juice
planning
information
competitivenessandfor supply
purchasing
and chain management,
decisions,
profitability, production,
providing impacting
stock
critical 2015).
segments,
the More the
refrigerated specifically
short
ones, in
shelf-life
such the and
as fresh
need
the foodto
dairy,industry,
maintain
fruit including
quality
and in
juice
competitiveness
information
information for
for and
purchasing
purchasing profitability,
decisions,
decisions, providing
production,
production, critical
stock
stock the refrigerated
segments,
segments, the
the ones,
short
short such
shelf-life
shelf-life as
and
and the
need
need dairy,
to
to fruit
maintain
maintain and
quality
qualityjuice
in
in
competitiveness
levels, logistics,
information for and profitability,
finance
purchasing and marketing
decisions, providing
(Yue et
production, al., critical
2016,
stock the refrigerated
storage
segments, theand ones,
distribution
short suchprocesses
shelf-life as the
and need dairy,
make
to fruit forecasting
sales
maintain and
qualityjuice
in
information
levels, for
logistics, purchasing
finance and decisions,
marketing production,
(Yue et al., stock
2016, segments,
the
the storage
storage theand
and short shelf-life
distribution
distribution and
processes need to
make maintain
sales quality
forecasting in
levels, logistics,
information finance
forMartínez
purchasing andal.,marketing
decisions, (Yue
production, 2016,
stock segments,
accuracy the short shelf-lifeprocesses for make
and need sales
to maintain forecasting
quality in
storagean andimportant
distributionfactor planning production,
Bertaglia, 2016, et 2018, Arvan et
et al., al.,
2018).
levels,
levels, logistics,
logistics,
Bertaglia,
Bertaglia, 2016,
2016,
finance
finance et
Martínez
Martínez
and
and
et al.,
al.,
marketing
marketing
2018,
2018, Arvan
Arvan
(Yue
(Yueet et
et 2018).
al., 2018). 2016, the
al., 2016,
al., the storage
accuracy
accuracy an
an
and distribution
important
important
processes
processes
factor
factor for
for
make
make
planning
planning
sales forecasting
sales production,
forecasting
production,
levels, logistics,
Bertaglia, 2016, finance and marketing (Yueet al.,
et al., 2016, the storage
minimizing
accuracy an and
lost distribution
sales
important due processes
to
factor a lack
for make
of sales
products,
planning forecasting
reducing
production,
Bertaglia,
In 2016, Martínez
the consumer Martínez
goods sector,
et
et al., 2018,
al.,where Arvan
Arvan et
2018, products al., 2018).
al.,
et are consumed accuracy
2018). minimizing
minimizing anlost
lost important
sales
sales due
due factor
to
to for of
a lack
a lack planning
of products,
products, production,
reducing
reducing
Bertaglia,
In the 2016,
consumer Martínez
goods et al.,
sector, 2018,
where Arvan
products et al.,
are 2018).
consumed accuracy
returns
minimizing due an to
lost important
the
sales proximity
due factor of for
the planning
expiration production,
date, and
In the
rapidly,consumer
sales goods
estimates sector, where
become products
even more are consumed
critical to minimizing
returns
returns due
due lost
to
to the
the due to
salesproximity
proximity to aaofof
lack
lack
the
the
of products,
ofexpiration
products, date,
expiration
reducing
reducing
date, and
and
In
In the
the consumer
consumer goods
goods sector,
sector, where
where products
products are
are consumed
consumed minimizing
improving
returns due lost sales
availability
to dueto to a lack
customers, of products,
thus reducing
reflecting on
rapidly,
rapidly,
In the sales
sales
consumer
business. Cohen
estimates
estimates
goods
et al.
become
become
sector,
(2017) where
cite
even
even
products more
more
fast-moving are critical
critical
consumed
consumer
to
to returns
improving
improving to the
the proximity
due availability
availability proximity
to
to
of
of the
customers,
customers, the expiration
expiration
thus
thus
date,
date, and
reflecting
reflecting and
on
on
rapidly,
rapidly, sales
sales estimates
estimates become
become even
even more
more critical
critical to
to returns
company
improving dueresultsto the
and
availability proximity
even to of
reducing
customers, the expiration
environmental
thus date,
damageand
business.
business.
rapidly,
goods
Cohen
Cohen
sales
(FMCG)
et al.
estimates
et
or al. (2017)
(2017)
become
Consumer
cite
cite fast-moving
fast-moving
even
Packaged more Goods
consumer
consumer
critical(CPG) to improving
company
company availability
results
results and
and even
even to reducing
customers,
reducing thus reflecting
environmental
environmental reflectingdamage
damage
on
on
business.
business.
goods Cohen
Cohen et
(FMCG) or
et al.
al. (2017)
(2017) cite
Consumer cite fast-moving
fast-moving
Packaged Goods consumer
consumer
(CPG) improving
when,
company availability
forresults
example, and even to
disposal customers,
reducing thus
by deterioration
environmental reflecting
is avoided
damage on
goods
business.
products, (FMCG)
Cohen
such as or Consumer
processed
et al. (2017)
foods, Packaged
cite fast-moving
beverages, Goods
canned (CPG)
consumer
goods, company
when,
when, for
for results
example,
example, and even
disposal
disposal reducing
by
by environmental
deterioration
deterioration is
is damage
avoided
avoided
goods
goods (FMCG)
(FMCG) or Consumer
or Consumer Packaged
Packaged Goods
Goods (CPG)
(CPG) when, company
(Doganis for results and even
al., 2005).disposal
et example, reducing
Ochiai (2015) by environmental
adds that the damage
accuracy
products,
products,
goods
soft drinks,
such
such
(FMCG) as
as
snacks,
processed
processed
orsweets
Consumer foods,
foods,
and
beverages,
beverages,
Packaged
chocolates,
canned
canned goods,
well asgoods,
asGoods items when,
(CPG) (Doganis
(Doganis foret example,
al., 2005).
2005). disposal
Ochiai
Ochiai by deterioration
(2015)
(2015) deterioration
adds
adds that
that the
the
is avoided
is accuracy
avoided
products,
products,
soft drinks, such
such as
as
snacks, processed
processed
sweets foods,
foods,
and beverages,
beverages,
chocolates, as canned
canned
well as goods,
goods,
items when,
in fresh
(Doganis for et
food
et example,
al.,
demand
al., 2005). disposal
forecasting
Ochiai byimproves
(2015) deterioration
adds the
that theis accuracy
avoided
efficiency
accuracy of
soft drinks,
products,
such as suchsnacks,
personal as sweets
processed
care and and
foods,chocolates,
cleaning beverages,
products. as well
canned
Due as
to items
goods,
rapid (Doganis
in
in fresh
fresh food
et
food al., 2005).
demand
demand Ochiai
forecasting
forecasting (2015) adds
improves
improves that
the
the the accuracy
efficiency
efficiency of
of
soft
soft drinks,
drinks, snacks,
snacks, sweets
sweets and
and chocolates,
chocolates, as well
as well as
as items
items (Doganis
order
in fresh and et
food inventory
al., 2005).
demand Ochiai
management,
forecasting (2015) adds
enabling that the
retailers accuracy
in this
such
such
soft as
as personal
personal
drinks,
deterioration, snacks,
many
care
care and
and
sweets
fast-moving
cleaning
cleaning
and products.
products.
chocolates,
consumer goods
Due
Due
as well have
to
to
as a
rapid
rapid
items
short in fresh
order
order and
andfood inventory forecasting improves
demandmanagement,
inventory management, improves
enabling
enabling
the
the efficiency
efficiency
retailers
retailers in
in
of
of
this
this
such
such as
as personal
personal
deterioration, many care
care and
and
fast-moving cleaning
cleaning products.
products.
consumer goods Due
Due
have to
toa rapid
rapid
short in fresh
segment
order andfood
to demand
reduce
inventory their forecasting
disposal
management, improves
volume by
enabling the
about efficiency
40%.
retailers in of
this
deterioration,
such
shelf as personal
life, as many
is fast-moving
care
the and
case consumer
cleaning
for goods products.
such goods
as have
Due
meats, toa short
rapid
fruits, order
segment
segment andto
to inventory
reduce
reduce their
theirmanagement,
disposal
disposal enabling
volume
volume by
by retailers
about
about 40%.
40%. in this
deterioration,
deterioration,
shelf life, as many
many
is thefast-moving
fast-moving
case for consumer
consumer
goods such goods
goods
as have a short
have afruits,
meats, order
short segment andto inventory
reduce theirmanagement,
disposal enabling
volume by retailers in this
shelf short segment to thereduce theirfactor
disposal volume by about
aboutArvan40%.
40%. et al.
shelf life,
shelf life, as
deterioration,
vegetables
life,
vegetables as
as
and
is
andmany the
dairy
isdairy
is the
case
thefast-moving
case
case
for
for goods
products,
for
products,
consumer
which
goods
goods
which
such
are goods
such
such
are
as
as meats,
highly
as
highly
have
meats,
meats,
afruits,
perishable.
fruits,
fruits,
perishable.
Regarding
segment to the
Regarding reduce human
humantheirfactor
disposal in the
in volume
the
predictions,
by aboutArvan
predictions, 40%. et al.
vegetables
shelf
Other life, and
as
products, dairy
isdairy
the as
such products,
case electronicswhich
for goods are
andsuch
fashionhighly
as meats, perishable.
apparel, fruits,
have (2018)Regarding
Regarding point the
the outhuman factor
that biases
human factor in
and
in the predictions,
systematic
the Arvan
errorsArvan
predictions, Arvan et
et al.
in demand
vegetables
vegetables
Other and
and
products, dairy
such asproducts,
products,
electronicswhich
which and are
are
fashionhighly
highly perishable.
perishable.
apparel, have Regarding
(2018) point the outhuman
that factor
biases in
and the predictions,
systematic errors in demand
et al.
al.
Other
short products,
vegetables and dairy
lifecycles, such as
rapidas electronics
products,
obsolescence,which and
are fashion
are highly
frequently apparel,
perishable.have
updated, (2018)
Regarding
forecasting
(2018) point
point the out
often
out that
human
occur
that biases
factor
in the
biases and
and systematic
insupply
the errors
predictions,
chain errors
systematic in
Arvan demand
decision-making
in et al.
demand
Other
Other
short products,
products,
lifecycles, such
such
rapid as electronics
electronics
obsolescence, and
and
are fashion
fashion
frequently apparel,
apparel, have
have
updated, (2018)
forecastingpoint out
often that
occur biases
in theand systematic
supply chain errors in demand
decision-making
short
Other
and lifecycles,
products,
have rapid
such
many competing asobsolescence,
electronics
alternatives. are
and frequently
fashion apparel,updated,
have (2018)forecasting
process
forecasting or often
point often occur
inoutsales
that
occur and in
biases
in theandsupply
operations chain
systematic
planningdecision-making
errors in demand
(S&OP), as
short
short
and
and
lifecycles,
lifecycles,
have
have many
many
rapid
rapid obsolescence,
competing
competing obsolescence,
alternatives.
alternatives.
are
are frequently
frequently updated,updated, process forecasting
process or
or often
in
in occurand
sales
sales and in the
the supply
supply chain
operations
operations chain
planning
planning
decision-making
decision-making
(S&OP),
(S&OP), as
as
short
and lifecycles,
have many rapid
competing obsolescence,
alternatives. are frequently updated, forecasting
influenced
process or byoften
in occur
personal
sales and in the
judgment. supply
operations chain
Puchalsky
planningdecision-making
et al. (2018)
(S&OP), add
as
and have many
Nagashima et al. competing
(2015) warn alternatives.
that increasing inventories due influenced process
influenced or by
by inpersonal
sales and
personal operations
judgment.
judgment. planning
Puchalsky
Puchalsky et al.(S&OP),
(2018)
(2018) as
add
add
and have
Nagashima manyet competing
al. (2015) alternatives.
warn that increasing inventories due process
that human
influenced or by in sales
decisions
personal and operations
become
judgment. planning
difficult
Puchalsky et
when
et al.
al.(S&OP),
appropriate
(2018) as
add
Nagashima
to the lack et
of aal. (2015)
more warn
accurate that
demand increasing
forecast inventories
has a due
greater influenced
that
that human
human by personal
decisions
decisions judgment.
become
become Puchalsky
difficult
difficult when
et
when al. (2018)
appropriate
appropriate add
Nagashima
Nagashima
to the lack et
of al. (2015)
(2015)
et aaal.more warn that
warn that
accurate demand increasing
increasing
forecast inventories
inventories
has due influenced
forecasting
due forecasting
aa greater that human by personal
models
decisions are judgment.
used
become Puchalsky
because
difficult they et
when (2018)
require
al. appropriate add
many
to
to
to
the
the
the
lack
Nagashima
impact lack
lack
of
on costs
of
of a
more
et aal.for(2015)
more
more
accurate
companies
accurate
accurate
demand
warn that
demand
demand deal forecast
increasing
with products
forecast
forecast
has
inventories
has
has aa such as that
greater
such due
greater
greater that
human models
forecasting
human
variables
decisions
models
decisions
to achieve
are
are become
used
used
become
greater accuracy,
difficultthey
because
because
difficult
pointing
when
they
when
appropriate
require
require
to the
many
many
appropriate
need for
impact
impact
to the on
on
lack costs
costs
of a for
for
morecompanies
companies
accurate that
that
demand deal
deal with
with
forecast products
products
has a such as
as
greater forecasting
forecasting
variables to models
models
achieve are
are
greater used
used because
because
accuracy, they
they
pointing require
require
to the many
many
need for
consumer
impact
impact on electronics,
costs for
on electronics, which
companies
costs for companies have
that a lower
deal
that aadeal with profit
products
withprofit
productsmargin such and
suchandas variables
forecasting
support
as support to
from achieve
models
automated greater
are accuracy,
used
tools. because pointing
they to the
require need
manyfor
consumer
consumer
impact on electronics,
costs for which
which
companies have
have
that lower
lower
deal with profit
productsmargin
margin such and
as variables
variables to
to achieve
from achieve
automated greater
greater accuracy,
accuracy, pointing
tools. pointing to to the
the need
need for
for
lose value
consumer
consumer quickly.
electronics,
electronics, Similarly,
which
which have
havethethe fashion,
aa lower
lower profit apparel
margin
margin and
profitapparel and support
and variables from automated
to achieve greatertools.
accuracy, pointing to the need for
lose value quickly. Similarly, fashion, support
support from automated tools.
lose
lose value
consumer
footwear
value quickly.
electronics,
segments,
quickly. amongSimilarly,
which have the
others,
Similarly, have
the fashion,
a lower profit
a short
fashion, apparel
margin and
life-cycle
apparel and Da Silvafrom
support fromet al. automated
(2017) tools.
automated comment on the use of forecasting
tools.
lose value
footwear quickly.
segments, amongSimilarly,
others, the
have fashion,
aa short apparel
life-cycle and Da
Da Silva
Silva et al. (2017) comment on the
the use
use ofof forecasting
forecasting
footwear
lose
high value
footwear segments,
demand quickly.
segments, among
volatility,
among others,
Similarly,
makingthe
others, have
have short
short life-cycle
it fashion,
difficult
adifficult apparel
for sales,
life-cycle and
and Da tools,
Da noting
Silva
Silva
et
et that(2017)
al.
al. time-series
(2017)
(2017)
comment
comment
comment methods on
on
on can use
the
the usually
use of
of be applied
forecasting
forecasting
footwear
high
high segments,
demand
demand among
volatility,
volatility, others,
making
making haveit
it a short
difficult life-cycle
for
for and
sales,
sales, tools,
tools, noting
notinget that
al.
that time-series
time-series methods
methods can
can usually
usually be applied
be applied
footwear
production
high segments,
demand and among
supply
volatility, chainothers, have
managers
making it atoshort life-cycle
accurately
difficult for and Da
predict
sales, to
tools,Silva
historical
notinget al. sales
that (2017)
data
time-seriescomment
with a high
methods on level
the
can use
of of
usually forecasting
precision,
be but
applied
high demand
production
production and
and volatility,
supply
supply chain
chain making
managers
managers it difficult
to
to accurately
accurately for sales,
predict tools,
to
to noting
historical
historical that
sales
sales time-series
data
data with
with methods
a high
aa high can
level
level usually
of
of be
precision,
precision, applied
but
but
high
sales demand
volumes
production and volatility,
(Nagashima
supply et
chain making
al., 2015,
managers it Yue difficult
to et al.,
accurately for predict
2016). sales, tools,
predict to noting
traditional
historical that
statistical
sales time-series
techniques
data with methods
are
high notcanthe
level usually
most
of be applied
appropriate
precision, but
production
sales
sales volumes
volumes and supply
(Nagashima chain
et managers
al., 2015, Yueto accurately
et al., 2016). predict to historical
traditional
traditional sales
statistical
statistical data with
techniques
techniques a high
are level
not the
arerelationship
not of
most
the of
most precision,
appropriate
appropriate but
production and(Nagashima
supply chain 2015, Yue
managers
et al., to accurately
et al., 2016). predict to historical
option
traditionalbecause sales
theydata
statistical with athehigh
consider
techniques level precision,
between but
input
sales
sales volumes
volumes (Nagashima
(Nagashima et al., 2015,
et al., 2015, YueYue et al., 2016).
et al., 2016). traditional
option because
option
traditionalbecausestatistical techniques
they consider
they
statistical consider the
techniques theare
are
not
not the
the most
arerelationship
relationship
not the
most
most
appropriate
appropriate
between
between input
input
appropriate
sales volumes (Nagashima et al., 2015, Yue et al., 2016). and
option output
because variables
they to be linear, which generally does not
option
and output
and because they consider
variables consider
to be linear, the
the relationship
linear, relationship
which generally between
between doesinput
input
not
and output
option
and
because
output
output
variables
variables
variables
to
to be
they consider
to be linear,
be linear, which
which generally
the relationship
which
between
generally
generally
does
does
not
doesinput
not
not
2405-8963 ©
Copyright © 2019,
2019 IFACIFAC (International Federation of Automatic Control) and output
752Hosting variables
by Elsevier Ltd. toAll be linear,
rights which generally does not
reserved.
Copyright
Peer review©
Copyright 2019
©under IFAC
2019 responsibility
IFAC 752
of International Federation of Automatic
752Control.
Copyright ©
© 2019
2019 IFAC
10.1016/j.ifacol.2019.11.203
Copyright IFAC 752
752
Copyright © 2019 IFAC 752
2019 IFAC MIM
738
Berlin, Germany, August 28-30, 2019 Elcio Tarallo et al. / IFAC PapersOnLine 52-13 (2019) 737–742

correspond to reality, suggesting the application of methods and methods. According to Bryman (2012), qualitative
based on Artificial Neural Networks. In fact, Akabane (2018) research generally emphasizes words rather than
points out that neural networks and Machine Learning are quantification in data collection and analysis.
part of the set of what are known as cognitive technologies,
which try to imitate human thinking and are able to On the other hand, the search for documentary and
manipulate large amounts of information in order to bibliographic sources is also essential to preventing a
objectively perform analyses, and can be applied to business duplication of efforts and to not repeat ideas that have already
processes. been expressed. Citations of conclusions from other authors
can highlight contributions of the research, evidence
Machine Learning is classified by Michalski, Carbonell and contradictions and confirm results (MarconI and Lakatos,
Mitchell (1983) into three forms: system engineering to learn 2003). The bibliometric approach, through which science is
from specific applications; an analysis of learning algorithms; portrayed by the results obtained, is based on the notion that
and a simulation of human learning processes. Mohri et al. the essence of scientific research is the production of
(2012) comment on the evolution of Machine Learning as a knowledge, and that scientific literature is a manifestation of
myriad of algorithms that can be divided into four main this knowledge (Okubo, 1997).
categories: i) supervised learning that assumes completely
labeled data in terms of input-output pairs to train an Regarding literature review, Creswell (2014) proposes that it
individual algorithm; ii) unsupervised learning that attempts provides a synthesis of the main studies on the research
to find structures in unlabeled data; iii) semi-supervised question, and can be used to indicate that the author is aware
learning operating on a mixture of labeled and unlabeled of the literature on the topic in question, as well as the latest
data; and iv) reinforcement learning based on the idea of publications.
maximizing a reward function by optimizing specific actions
in a given parametric configuration. 3. LITERATURE REVIEW

According to Goodfellow, Bengio and Courville (2015), The literature review performed in this research has identified
Deep Learning is a more complex form of machine learning relevant papers related to Machine Learning applied to
that allows computers to learn from experience without the demand prediction for fast-moving consumer goods, focusing
need for knowledge specification by a human operator, on the benefits achieved, business sectors addressed and the
understanding the world in terms of a hierarchy of concepts possible advantages over traditional statistical techniques.
with deep layers built on top of one another. The importance of sales forecasting is shown in older
Thus, the question to be answered by this study is: What are publications (2004 and 2005), as well in recent surveys (from
the benefits of applying Machine Learning to demand 2016 to 2018), which mention the quality of forecasts as a
forecasting for manufacturers and retailers of fast-moving critical success factor with a large impact on business
consumer goods? The objective is to review scientific management, including production, supply chain, logistics
literature and identify if there are advantages over traditional and stores. A bibliometric analysis of articles published from
statistical techniques, what gains are obtained as a result of 2011 to 2017, searched in the Scopus and Web of Science
higher sales predictability, and which business segments are databases, regarding Machine Learning for sales prediction,
addressed. shows an increase in the number of publications between
2014 and 2016, with a decrease in 2017, as shown in Fig. 1.
2. METHODOLOGY It represents 32 articles with 314 citations.

This is an exploratory research with a qualitative approach,

carried out through bibliographical research and literature
review, seeking the benefits of Machine Learning applied to
demand forecasting for fast-moving consumer goods.
This technique was chosen in order to achieve more
comprehensive information in this study, and to allow for a
more exhaustive study of specific topics in future research.
Exploratory research provides an overview and
approximation on a theme, constituting the first stage of a Fig. 1. Recent history of articles on Machine Learning in
broader investigation, allowing the definition of the theme, demand forecasting for fast-moving consumer goods
delimitation of the research scope, formulation of hypotheses
or discovery of a new approach (GIL, 2008, PRODANOV, One can observe in the literature that statistical techniques
2013). and tools have been used by companies to estimate demand
for their products, using historical sales series as a main data
Flick (2009) proposes that the essential attributes of source. However, according to Kandananond (2012), demand
qualitative research are choosing appropriate methods and forecasts with Machine Learning have had higher quality
theories, the identification and analysis of different results than traditional techniques, including known
perspectives, reflection on the research as part of the approaches such as the Autoregressive Integrated Moving
knowledge production process, and the variety of approaches Average (ARIMA) model, especially for consumer goods and

753
 2019 IFAC MIM
Berlin, Germany, August 28-30, 2019 Elcio Tarallo et al. / IFAC PapersOnLine 52-13 (2019) 737–742 739

complex product hierarchies, presenting smaller error sales, promotions, product characteristics, climate and
deviations. Wu and Zheng (2015) present sales forecasting economic indicators, resulting in greater precision in sales
models through Machine Learning, achieving greater forecasting and better replenishment at retail stores, thereby
accuracy than traditional statistical models for products with improving supply chain efficiency. Lu (2014) confirms the
highly volatile demand and very short life cycles, such as in need to include new variables in the forecast models, stating
fast fashion retail, especially when data other than sales that many factors, even stock market indexes, affect the
history are added to the model - in this case, information on accuracy of sales forecasting, and that the selection of
product searches over the internet was included. Tsoumakas variables is therefore crucial. Several data sources were used
(2018) states that Machine Learning techniques are more by Qu et al. (2017), including both internal data (sales
effective and flexible than traditional statistical techniques for history, discounts granted and inventories levels) and external
predicting time series because they have greater processing variables (holidays and regional factors) in their Machine
power and the ability to handle additional variables, listing Learning algorithms, with the main results being a better
those that he considers most appropriate for time series sales forecast and an adjustment of the stock levels of semi-
forecasting: simple ones, like Moving Average; hybrid luxury goods stores, which have seasonal characteristics and
approaches with a Radial Basis Function network; fuzzy large variations in purchase stimulus.
algorithms; Genetic Algorithms; joint approaches
(ensembles) that use learning algorithms like decision tree In the technology sector, Lu and Shao (2012) have acquired
learners, rule learners, lazy learners, Support Vector more precise sales forecasts for short life-cycle products
Machines, Logistic Regression and, more recently, Deep through Extreme Learning Machine algorithms, enabling an
Learning as the Stacked Denoising Autoencoder. improvement of the refueling process of a computer products
and accessories retail chain, whose main feature is
Several Machine Learning techniques are presented by the technological obsolescence. Lu (2014) used a Support Vector
authors to obtain better sales forecasts. Chen and Ou (2011) Machine (SVM) to improve the sales management
applied Extreme Learning Machine networks combined with effectiveness of computing products that are highly
other methods to real sales data and weather forecasting, replaceable and subject to dramatic changes in demand. A
demonstrating an improvement in retail sales forecast predictive sales model for innovative products was developed
accuracy with the objective of making better production by Lee, Kim, Park and Kang (2014) in a study on 3D TVs,
decisions, increasing revenue, reducing losses on unsold whose technology had no sales history, by using a joint
products, and increasing customer satisfaction through an approach of Machine Learning techniques, which resulted in
increased availability of items at points of sale. Islek and greater accuracy than other techniques available. For
Oguducu (2015) took another approach to developing a information technology goods, Lu and Chang (2014) have
demand forecasting methodology based on Bayesian developed a hybrid sales prediction model, which includes
Networks, a technique that verifies conditional probabilities SVM, with greater accuracy and reliability than previous
according to the data provided, resulting in improved solutions. Chen and Lu (2017) confirm the benefits of
accuracy of regional sales forecasts for a wide variety of Machine Learning for technology businesses with short life
retail products, distributed through warehouses of various cycle products, by improving a computer retailer’s sales
sizes. In online retailing, Ponce, Miralles and Martinez forecasting accuracy and inventory management.
(2015) compared traditional Machine Learning techniques
with Artificial Hydrocarbon Networks, which is a supervised Studies related to the fashion industry reveal the search for
learning technique, showing the potential for improving sales better and faster demand forecasting. Yu, Choi, and Hui
estimate quality. New technologies such as image recognition (2011) met an online fashion product store’s business
methods were applied by Wang et al. (2015) in order to objective of having faster and more frequent sales forecasts
predict consumer buying decisions based on the collected for a large volume of products (SKUs) by applying Extreme
brain images of a group of people when subjected to pictures Learning Machine (ELM) techniques to historical sales data
of products, through Support Vector Machines (SVM), which with the volumes and characteristics of the items sold. Liu et
are supervised learning models with associated learning al. (2013) show the evolution of sales forecasts in fashion
algorithms. Kaneko and Yada (2016) achieved high-level retail, highlighting ELM techniques and benefits, such as
precision in the sales forecast for a retail stores chain by proper inventory management. Choi (2014) has developed an
applying Deep Learning techniques on three-year daily data intelligent ELM algorithm for fashion retail sales
from points of sale, which ultimately led to the definition of a management, attaining greater precision and speed in the
new management model with improvements in inventory calculation of estimates. Tehrani and Ahrens (2016) reduced
levels, product categorization and distribution. the sales forecast discrepancy rate for fashion items with a
high production scale, thereby reducing write-off losses and
Some authors emphasize the importance of data variables. increasing profitability through Machine Learning algorithms
According to Guo, Wong, and Li (2013), a variety of factors on historical and catalog sales data. Ferreira, Lee and Simchi-
influence a product’s sales, such as its own attributes and the Levi (2016) showed revenue increasing for an online fashion
economic environment, and therefore there is no pattern in clothing and accessories store, where Machine Learning
the time series data used for estimates. As an alternative for examined the lost sales history and predicted future demand
solving this problem, they presented a sales prediction model for new products (launches), contributing to the daily pricing
based on Machine Learning techniques with multiple process and following market dynamics. Teucke et al. (2016)
variables, composed of data from early orders, historical utilized a combination of methods in their prediction model,

754
2019 IFAC MIM
740
Berlin, Germany, August 28-30, 2019 Elcio Tarallo et al. / IFAC PapersOnLine 52-13 (2019) 737–742

including fuzzy logic, with the direct benefit of more accurate environment of simultaneous and concurrent promotional
sales forecasts for seasonal fashion products. These forecasts actions. Their study analyzed a seasonal product (beer) and a
were also faster than with previous methods, and were done non-seasonal item (milk). Similarly, Pinho, Oliveira and
in time to make production decisions, which made it possible Ramos (2016) have achieved better sales forecast accuracy
to reduce the rate of inventory shortages and excesses. Ren, levels in the categories of beverages and cleaning products,
Chan and Ram (2017) state that in the fashion industry where analyzing the sales promotions carried out at a retail stores
the products are manufactured, consumed and disposed of in chain.
a short period of time, demand is highly volatile, and the
inventory is distributed, Machine Learning for sales 4. ANALYSIS AND DISCUSSION
forecasting has shown positive results via greater speed and
reliability. In the context of the fast-moving consumer goods (FMCG)
industry, most of the papers from this research investigation
In the case of short shelf-life foods, Agrawal and Schorling show that Machine Learning results in better demand
(1996, apud Chen and Ou, 2009) developed an Artificial predictability than the use of traditional models. One of the
Neural Network (ANN) prediction model for a perishable and benefits obtained from this better sales forecast accuracy is an
refrigerated food convenience stores network that reached improvement in inventory management due to better
higher accuracy than traditional techniques, and noted that decisions on production and distribution. The decrease in the
demand forecasting accuracy plays a crucial role in the occurrence of stockouts, which cause a shortage of items in
profitability of retail operations, especially for food preserved stores, is also mentioned, as well as the greater availability of
at low temperatures. In this sense, Da Silva et al. (2017) products at the points of sale, thus increasing revenue and
noted that using Support Vector Machine (SVM) for demand customer satisfaction.
forecasting, combined with other techniques, led to a
reduction in losses from unsold dairy products that had The main advantages cited over traditional statistical
passed their expiration date. Ochiai (2015) has developed techniques are better predictions and more flexibility,
algorithms with Heterogeneous Mixture Learning technology especially when new data variables are included in the
to estimate the demand of a short-term food grocery store estimation models, in addition to the commonly used
chain, with a significant reduction in the disposal of unsold historical sales series, increasing data volume and analysis
items. complexity. The significant increase in the number of internal
and external variables, the growing data volume of time
Regarding the food and beverage industry, Fujimaki et al. series and the complexity of relationships between the factors
(2016) built a model that aggregates Machine Learning for that can influence sales all encourage the adoption of
demand forecasting and price-optimization techniques for a Artificial Neural Networks and Machine Learning.
beverage retailing company, which resulted in better
forecasts, greater reliability for decision makers, and, as a The market sectors cited by the articles researched can be
result, a revenue increase estimated at 16%. Tsoumakas seen in Fig. 2, which shows the distribution and highest
(2018) shows the advantages of Machine Learning techniques incidence of study (51%) for the fashion and technology
for sales prediction in the food industry, such as in segments.
supermarkets, grocery stores, restaurants, bakeries and
confectioneries, where accurate short-term forecasting allows
for inventory level minimization, the elimination of expired
products at stores and, at the same time, prevents the loss of
sales due to stockout. The main benefits observed were the
reduction of human bias due to the greater automation of the
forecasting process, a higher degree of forecast precision, and
the flexibility to change variables. Drawbacks for companies
that the author mentioned are the possible unavailability of
detailed historical information and the large number of
learning algorithms available, which could make choosing
the proper one difficult. Fig. 2. Article distribution by market segments
A common aspect among the various business sectors, as
More specific benefits are seen in some particular business
highlighted by Cohen et al. (2017), is the importance of sales
areas. In the fresh food sector, more accurate forecasts lead to
promotions, which are actions encouraging customers to buy
a reduction in loss from products that reach their expiration
fast-moving consumer products (FMCG). This is due to
dates, and costs of transportation and storage of refrigerated
promotional expenses and the fact that a large part of FMCG
products can be reduced. In the fashion business, one of the
products are sold as promotions: 12% to 25% of supermarket
challenges is to prevent losses due to unsold products
sales in five European countries are made in this form,
resulting from constant changes in consumer trends and
according to a 2014 Nielsen survey. Soguero et al. (2012)
preferences: Machine Learning techniques have proved
propose an estimating model for promotional sales using
adequate for both better predictability and faster processing
Machine Learning, with benefits through greater promotional
speed, capable of generating estimates according to the
efficiency and predictability in a dynamic and complex
frequency of new collections. Manufacturers and retailers of

755
 2019 IFAC MIM
Berlin, Germany, August 28-30, 2019 Elcio Tarallo et al. / IFAC PapersOnLine 52-13 (2019) 737–742 741

technology products, which can quickly be replaced by new available, thereby making the choice for the best alternative
releases, benefit from better sales and inventory management. difficult. New research on real implementation cases and
In addition, better predictions were obtained for innovative adopted techniques could help with that issue.
products that do not have historical data.
Thus, this study contributes to the identification of the
Finally, the research shows that promotional actions are Machine Learning benefits and characteristics, applied in
exceptionally important for all sectors, due to the significant improving demand forecasting accuracy in the FCMG
share of consumer goods sales, but make the task of creating industry, which is a key element of competitiveness.
estimates more complex thanks to the different possible
combinations of product types, categories, periods, customers REFERENCES
and regions, as well the impact of items not participating in
Akabane, G. (2008). Gestão estratégica das tecnologias
promotions. Recent studies in 2012, 2014, 2016 and 2017
present the advantages of Machine Learning in improving cognitivas: conceitos, metodologias e aplicações.
promotional efficiency, with greater predictability and better Saraiva, São Paulo.
Arunraj, N. S., Ahrens, D. (2015). A hybrid seasonal
results.
autoregressive integrated moving average and quantile
regression for daily food sales forecasting. International
5. CONCLUSIONS
Journal of Production Economics, Vol.170, 321-335.
The results of this research show that Machine Learning Arvan, M., Fahimnia, B., Reisi, M., Siemsen, E. (2018).
techniques, including the latest methods such as Deep Integrating Human Judgement into Quantitative
Learning, as well as utilizing a combination of various Forecasting Methods: A Review. Omega.
techniques applied in demand forecasting models, can Bertaglia, P. R. (2016). Logística e gerenciamento da cadeia
provide benefits to manufacturers and retailers of fast-moving de abastecimento. 3ª ed. Saraiva, 2016.
consumer goods. Bryman, A. (2012). Social Research Methods. 4th ed. Oxford
University Press.
The main benefit observed was a higher level of precision in Chen F.L., Ou T.Y. (2011). Sales forecasting system based
demand forecasting. As a result, manufacturers can build on Gray extreme learning machine with Taguchi method
better sales and operations plans (S&OP), adjust production in retail industry. Expert Systems with Applications.
volumes and regional distribution, and improve inventory Chen I.-F., Lu C.-J. (2016). Sales forecasting by combining
balancing throughout the chain. Retailers can be more clustering and machine-learning techniques for computer
efficient in store management, restocking processes, and retailing. The Natural Computing Applications Forum.
maintaining adequate supply with fewer shortages or Choi, Tm; Hui, Cl., Liu, N., Ng, Sf., Yu, Y. (2014). Fast
excesses, thereby resulting in cost savings, increased revenue, fashion sales forecasting with limited data and time.
and greater customer satisfaction through the availability of Decision Support Systems.
desired products. Cohen, M. C., Leung, N-H. Z., Panchamgam, K., Perakis, G.,
Smith, A. (2017). The Impact of Linear Optimization on
Comparisons with traditional statistical techniques showed
Promotion Planning. Operations Research.
better sales forecasting, as demonstrated by higher accuracy Creswell, J.W. (2014). Research Design. Qualitative,
than previous models, the flexibility to handle a greater Quantitative, and Mixed Methods Approaches. 4th ed.
number of data variables and the capacity to process large Sage.
data volumes. Studies showed that complex analysis, such as Da Silva, I.D., Moura, M.D.C., Didier Lins, I., López
the impact of promotions, are best handled by Machine Droguett, E., Braga, E. (2017) Non-Stationary Demand
Learning solutions. Forecasting Based on Empirical Mode Decomposition
The most often mentioned business segments are fashion, and Support Vector Machines. IEEE Latin America
especially what is known as fast-fashion, technology Transactions, 15 (9), art. no. 8015086, 1785-1792.
consumer products, general retail, short shelf-life foods, and Doganis, P., Alexandridis, A., Patrinos, P., Sarimveis, H.
food & beverage. Some sector-specific business challenges (2005). Time series sales forecasting for short shelf-life
addressed were the need for more frequent forecasts, food products based on Artificial Neural Networks and
prediction for innovative products with no sales history and evolutionary computing. Journal of Food Engineering,
the reduction of food losses by expiration date, which may Vol.75(2), 196-204.
contribute to sustainability in the fresh food sector. Ferreira K.J., Lee B.H.A., Simchi-Levi D. (2016) Analytics
for an online retailer: Demand forecasting and price
Due to the increasing digitalization of the retail sector, future optimization. Institute for Operations Research and the
research could explore Deep Learning capabilities to handle Management Sciences, v. 18, n. 1, 69-88.
different type of data such as images, data from Internet of Flick, U. (2009). An Introduction to Qualitative Research. 4th
Things (IoT) sensors and consumer behavior information, ed. Sage.
enabling better demand predictions and even decisions in real Fujimaki R., Muraoka Y., Ito S., Yabe A. (2016). From
time that could affect the entire supply chain. prediction to decision making - Predictive optimization
technology. NEC Technical Journal.
One concern regarding adopting Machine Learning for Gil, A.C. (2008). Métodos E Técnicas De Pesquisa Social. 6ª
demand prediction could be the wide variety of algorithms ed. Editora Atlas, São Paulo.

756
2019 IFAC MIM
742
Berlin, Germany, August 28-30, 2019 Elcio Tarallo et al. / IFAC PapersOnLine 52-13 (2019) 737–742

Goodfellow, I., Bengio, Y and Courville, A. (2015). Deep Pinho, JM., Oliveira, JM; Ramos, P. Sales. (2016).
Learning. Cambridge: The MIT Press. Forecasting in Retail Industry Based on Dynamic
Guo, Z.X., Wong, W.K., Li, M. (2013). A multivariate Regression Models. Advances in Manufacturing
intelligent decision-making model for retail sales Technology. IOS Press, 483.
forecasting. Decision Support Systems, v. 55, n. 1, 247- Ponce, H., Miralles-Pechuan, L., Martinez-Villasenor, MD.
255. (2015). Artificial Hydrocarbon Networks for Online
Islek, I., Oguducu, Sg. (2015). A Retail Demand Forecasting Sales Prediction. Advances in Artificial Intelligence and
Model Based on Data Mining Techniques. IEEE 24th Its Applications, MICAI 2015, PT II.
International Symposium on Industrial Electronics Prodanov, C.C. (2013). Metodologia do Trabalho Científico:
(ISIE). Métodos e Técnicas da Pesquisa e do Trabalho
Kandananond, K. (2012). A comparison of various Acadêmico. 2ª ed. Universidade Feevale, Novo
forecasting methods for autocorrelated time series, Hamburgo, RS.
International Journal of Engineering Business Puchalsky, W., Ribeiro, Gl T., Da Veiga, C P., Freire, R Z.,
Management. Santos Coelho, L Dos. (2018). Agribusiness time series
Kaneko, Y., Yada, K. (2016). A Deep Learning Approach for forecasting using Wavelet neural networks and
the Prediction of Retail Store Sales. IEEE 16th metaheuristic optimization: An analysis of the soybean
International Conference on Data Mining Workshops sack price and perishable products demand. International
(ICDMW). Journal of Production Economics, Vol.203, 174-189.
Lee H., Kim S.G., Park H.-W., Kang P. (2014). Pre-launch Qu T., Zhang J.H., Chan F.T.S., Srivastava R.S., Tiwari
new product demand forecasting using the Bass model: M.K., Park W.-Y. (2017). Demand prediction and price
A statistical and machine learning-based approach. optimization for semi-luxury supermarket segment.
Technological Forecasting & Social Change, Vol.86, Computers & Industrial Engineering.
49-64. Ren, SY; Chan, HL; Ram, P. (2017). A Comparative Study
Liu, N; Ren, Sy; Choi, Tm; Hui, Cl; Ng, Sf. (2013). Sales on Fashion Demand Forecasting Models with Multiple
Forecasting for Fashion Retailing Service Industry: A Sources of Uncertainty. Annals of Operations Research,
Review. Mathematical Problems in Engineering. Spring.
Lu, C.-J. (2014). Sales forecasting of computer products Soguero-Ruiz C., Gimeno-Blanes F.-J., Mora-Jiménez I.,
based on variable selection scheme and support vector Martínez-Ruiz M.P., Rojo-Álvarez J.-L. (2012). On the
regression. Neurocomputing, n. 128, 491-499. differential benchmarking of promotional efficiency with
Lu, CJ., Chang, CC. (2014). A Hybrid Sales Forecasting machine learning modelling (II): Practical applications.
Scheme by Combining Independent Component Expert Systems with Applications.
Analysis with K-Means Clustering and Support Vector Tehrani, Af., Ahrens, D. (2016). Enhanced predictive models
Regression. Scientific World Journal. for purchasing in the fashion field by using kernel
Lu C.-J., Shao Y.E. (2012). Forecasting computer products machine regression equipped with ordinal logistic
sales by integrating ensemble empirical mode regression. Journal of Retailing and Consumer Services.
decomposition and extreme learning machine. Teucke, M., Ait-Alla, A., El-Berishy, N., Beheshti-Kashi, S.,
Mathematical Problems in Engineering. Lutjen, M. (2016). Forecasting of Seasonal Apparel
Marconi, M.A.; Lakatos, E.M. (2003). Fundamentos Da Products. Dynamics in Logistics, LDIC, Springer
Metodologia Científica. 5ª ed. Editora Atlas, São Paulo. International Publishing AG.
Martínez, A., Schmuck, C., Pereverzyev, S., Pirker, C., Tsoumakas, G. (2018). A survey of machine learning
Haltmeier, M. (2018). A machine learning framework for techniques for food sales prediction. Artificial
customer purchase prediction in the non-contractual Intelligence Review – Springer, Netherlands.
setting. European Journal of Operational Research. Wang Y., Chattaraman V., Kim H., Deshpande G. (2015).
MichalskI, R. S., Carbonell, J. G. and Mitchell, T. M. (1983). Predicting Purchase Decisions Based on Spatio-
Machine Learning. Tioga publishing company, Palo Temporal Functional MRI Features Using Machine
Alto. Learning. IEEE Transactions on Autonomous Mental
Mohri, M., Rostamizadeh, A., Talwalkar, A. (2012). Development.
Foundations of Machine Learning. The MIT Press. Wu J., Zheng S. (2015). Forecasting for fast fashion products
Nagashima, M., Wehrle, F. T., Kerbache, L., Lassagne, M. based on web search data by using OS-ELM algorithm.
(2015). Impacts of adaptive collaboration on demand Journal of Computational Information Systems, 11(14),
forecasting accuracy of different product categories 5171-5180.
throughout the product life cycle. Supply Chain Yu Y., Choi T.-M., Hui C.-L. (2011). An intelligent fast sales
Management, Vol.20(4), 415-433. forecasting model for fashion products. Expert Systems
Ochiai, K. (2015). Predictive analytics solution for fresh food with Applications.
demand using heterogeneous mixture learning Yue, L., Wangwei, J., Jianguo, Z., Junjun, G., Jiazhou, Z.,
technology. NEC Technical Journal, v.10, n.1, 83-86. Aiping, J. (2016). Product life cycle based demand
Okubo, Y. (1997). Bibliometric Indicators and Analysis of forecasting by using artificial bee colony algorithm
Research Systems. Methods and Examples. Oecd optimized two-stage polynomial fitting. Journal of
Science Technology and Industry Working Papers.Oecd Intelligent & Fuzzy Systems, Vol.31 (2), 825-837.
Publishing, Paris.

757