Demand forecasting for apparel
manufacturers by using
neuro-fuzzy techniques
Asli Aksoy
Industrial Engineering Department, Faculty of Engineering and Architecture,
Uludag University, Bursa, Turkey and
Faculty of Social Sciences, Economics and Business Administration,
University of Bamberg, Bamberg, Germany
Nursel O
¨ztu
¨rk
Industrial Engineering Department, Faculty of Engineering and Architecture,
Uludag University, Bursa, Turkey, and
Eric Sucky
Faculty of Social Sciences, Economics and Business Administration,
University of Bamberg, Bamberg, Germany
Abstract
Purpose – According to literature research and conversations with apparel manufacturers’
specialists, there is not any common analytic method for demand forecasting in apparel industry
and to the authors’ knowledge, there is not adequate number of study in literature to forecast the
demand with adaptive network-based fuzzy inference system (ANFIS) for apparel manufacturers.
The purpose of this paper is constructing an effective demand forecasting system for apparel
manufacturers.
Design/methodology/approach – The ANFIS is used forecasting the demand for apparel
manufacturers.
Findings – The results of the proposed study showed that an ANFIS-based demand forecasting
system can help apparel manufacturers to forecast demand accurately, effectively and simply.
Originality/value – ANFIS is a new technique for demand forecasting, combines the learning
capability of the neural networks and the generalization capability of the fuzzy logic. In this study, the
demand is forecasted in terms of apparel manufacturers by using ANFIS. The input and output
criteria are determined based on apparel manufacturers’ requirements and via literature research and
the forecasting horizon is about one month. The study includes the real-life application of the proposed
system, and the proposed system is tested by using real demand values for apparel manufacturers.
Keywords Decision making, Forecasting
Paper type Research paper
1. Introduction
The term supply chain is used to describe the flow of goods from the very first process
encountered in the production of a product right through to the final sale to the end
consumer (Bruce et al., 2004). There are many factors affect supply chain performance.
The current issue and full text archive of this journal is available at
www.emeraldinsight.com/1746-5664.htm
The authors would like to thank the anonymous referees for reading the paper and offering many
helpful comments.
Received 5 October 2011
Revised 16 January 2012
4 April 2012
29 August 2012
Accepted 17 September 2012
Journal of Modelling in Management
Vol. 9 No. 1, 2014
pp. 18-35
qEmerald Group Publishing Limited
1746-5664
DOI 10.1108/JM2-10-2011-0045
JM2
9,1
18
One of the most important factor also affects the planning decisions of the other
departments is the accuracy of forecast. Because most retailers do not know their
demand with certainty, they have to make their inventory decisions based on demand
forecasts. With inaccurate forecasts, the quantity of materials ordered does not match
the demand. Inaccurate forecasts can, therefore, significantly influence the performance
of the supply chain in terms of increased inventory costs, backorders or loss of sales, and
customer goodwill throughout the supply chain. They can also cause low utilization of
capacity and other problems in production (Zhao et al., 2002). Even with much effort and
funds spent to improve supply chain processes, they still lack reliability and efficiency if
the demand forecast accuracy is poor (Sayed et al., 2009).
Forecasting refers to computing the probability of the future value. The underlying
assumption in most forecasting methods is that the past patterns or behavior will
continue in the future (Frank et al., 2003). The ability to respond to customer
requirements on a timely basis has always been a fundamental element of the marketing
concept. However, there has never been as much pressure as there is today to further
accelerate the responsiveness of marketing systems (Christopher and Peck, 1997).
In the present world, all industries need to be adaptable to a changing business
environment in the context of a competitive global market. To make decisions related
to the conception and the driving of any logistic structures, industrial managers must
rely on efficient and accurate forecasting systems (Sun et al., 2008). A good forecasting
system is essential for avoiding problems such as inventory shortages and excesses,
missed due dates, plant shutdowns, lost sales, lost customers, expensive expediting,
and missed strategic opportunities (Frank et al., 2003). Improving forecasting models is
considered a vital part in the overall supply chain process (Sayed et al., 2009).
In this research, an adaptive network-based fuzzy inference system (ANFIS) is used
to forecast the demand for the apparel industry. Inputs and outputs of the system,
training data set, and system rules are determined by interviewing specialists in
apparel firms and via literature research. The remainder of this paper is organised as
follows. Section 2 presents a literature review of the apparel industry and demand
forecasting in supply chain management. Section 3 explains the proposed approach
and the demand forecasting model for an apparel manufacturer based on the ANFIS.
The application examples and results are provided in Section 4. Finally, conclusions
are presented in Section 5.
2. Literature review
2.1 Apparel industry
Apparel industry is a major sector for both the industrialised and the lesser developed
economies, contributing both to wealth generation and employment (Bruce et al., 2004).
The year 2005 marked the end of a 30-year stretch of textile quotas. As the industry
enters the global “free market” competition, many developing countries fear the
uncertainty in the existing competition (Ballestero, 2004). To enhance the commercial
competitive advantage in a constantly fluctuating environment, apparel companies
must improve their supply chain management, which requires sales forecasting
systems adapted to the uncertain environment of the apparel industry. The accuracy of
the demand forecast significantly affects inventory levels, costs and customer
satisfaction levels. To set up all of the logistic steps required for producing and selling
a product, apparel managers must rely on efficient and accurate forecasting systems
Demand
forecasting
19
(Thomassey et al., 2002). The effectiveness of the supply chain optimisation depends on
the forecast accuracy of the finished product sales (Graves et al., 1998).
The fashion industry is characterised by aggressive competition levels, a large
marketplace sourcing high-variety, high-margin, short product life cycles in a global
context, high level of impulse purchasing and unpredictable demand (Bruce et al., 2004;
Mason et al., 2007; Cao et al., 2008). The short life cycle of products implies that
available sales data are reduced (Thomassey et al., 2002). Demand forecasting in
apparel industry is very complex. Indeed, a wide range of textile item references exists,
and their historic sale data are often short and particularly perturbed by numerous
factors, which are neither strictly controlled nor identified (DeToni and Meneghetti,
2000). These factors can depend on the item (colours, prices, etc.), distributor (number
of stores, merchandizing, etc.), customers (fashion, etc.) or external factors (weather,
holidays, etc.). These factors have different impacts on sales and not always available.
In the apparel industry, the lead times from retailer order to delivery are quite long.
As a result of these long lead times, the risks of having too little or too much inventory
increases if the retailers have to place the orders long before the season. In addition to
long lead times, product proliferation is increasing the risks a manufacturer faces. As the
product proliferation increases, the variability of demand for each time also increases
(Tan, 2001). The long supply pipeline makes the lead time of textile-apparel supply chain
relatively long and uncertain in response to the volatile characteristics of fashion markets.
So, the coordination in textile-apparel supply chain becomes even more important
(Cao et al., 2008). In order to reduce stocks and to limit stock out, apparel companies require
specific and accurate demand forecasting systems (Thomassey et al., 2005).
The last decade apparel industry has learnt that the supply chain can be made
more efficient in order to decrease lead times and related demand forecast errors
(Jacobs, 2006). A demand forecasting system is required to respond to the versatile
fashion market and the needs of the distributors. Nowadays, due to the specific constraints
of the apparel sales (numerous and new items, short life time), existing forecasting models
are generally unsuitable or unusable (Thomassey and Fiordaliso, 2006).
An apparel demand forecasting system requires; to quickly react to a significant
variation of trend and seasonality, to identify and to smooth purely random events, to
perform forecasting on short historic sales data and to take into account the influences
of explanatory variables such as: product features, marketing strategy, distribution
area, distribution mode, competitive environment, space-time environment, consumer
environment, economic situation (Thomassey et al., 2002).
2.2 Forecasting in supply chain management
To have an available decision-making system is becoming a crucial issue for
organisations in a constantly fluctuating environment where the economic uncertainty
needs mathematical models. Forecasting the expected demand for a certain period of
time with one or more products is one of the most relevant targets in an enterprise.
Despite the need for accurate forecasting to enhance the commercial competitive
advantage, there is no standard approach (Efendigil et al., 2009).
In practice and in literature, various demand forecasting techniques have been
studied and used. Most of these techniques are based on statistical methods such as
moving average, time series analysis, exponential smoothing, Box-Jenkins method, and
casual models. These methods assume that historical data are recorded in the past and
JM2
9,1
20
that they are precisely known. Furthermore, the statistical forecast methods assume
that a historical pattern of demand is a good indicator of future demand. They can be
applied successfully when historical data are reliable and when environments being
forecasted are relatively stable. However, these methods are perceived as being slow to
react to changes in dynamic environments (Petrovic et al., 2006).
Kuo et al. (2002) stated that statistical methods, such as regression modelling and
ARIMA, have been the candidates for decision makers; however, these methods are only
efficient for data that are seasonal or cyclical. If the data are influenced by a special case,
such as a promotion, these methods are not feasible. Efendigil et al. (2009) also concluded
that statistical methods are only efficient for data having seasonal or trend patterns,
while artificial neural techniques are also efficient for data that are influenced by special
cases, such as promotions or extreme crises. Huang (2009) presented that statistical
methods frequently fail to accurately capture and manage the components of random
variability in demand.
Garetti and Taisch (2009) determined some limitations using quantitative methods.
First, a lack of expertise might cause a misspecification of the functional form linking
the independent and dependent variables together, resulting in a poor regression.
Second, a large amount of data is often required to guarantee an accurate prediction.
Third, non-linear patterns are difficult to capture. Finally, outliers can bias
the estimation of the model parameters. Some of these limitations can be overcome
by the use of neural networks (NNs), which have been mathematically demonstrated to
be universal approximators of functions.
A literature review reveals no effective and practicable forecasting method for
forecasting demand for a range of products with demand having high random
volatility (Huang, 2009). Escoda et al. (1997) stated that the uncertainty in economic
environment makes the design of mathematic models with statistical methods very
difficult. Thus, if the economic environment is uncertain, fuzzy logic may help to solve
problems that are difficult to solve by the use of traditional methods.
Escoda et al. (1997) compared the NN, fuzzy NN and Winter’s method for demand
forecasting. They had 73 historical data points and used 59 of them in the training stage
and 14 in the validation stage. They stated that the fuzzy NN system generates more
accurate results than do the other two methods. Chen et al. (2000) focused on determining
the impact of demand forecasting on the bullwhip effect. They have shown that if a
retailer periodically updates the mean and variance of demand based on observed
customer demand data, then the variance of the orders placed by the retailer will be
greater than the variance of demand. They have also shown that providing each stage of
the supply chain with complete access to customer demand information can
significantly reduce the increase in variability. Frank et al. (2003) forecasted demand
by using three different methods: exponential smoothing, Winter’s method and NN
using the same historical data. They stated that the forecasting accuracy of the NN
method is higher than those of the other two methods. Thomassey et al. (2005) developed
a specific demand forecasting tool for the textile market. Their method was based on two
different models: the FIS for the median-term forecasting model and the neuro-fuzzy
system for the short-term forecasting system. They stated that the use of the FIS results
in more accurate forecasts than with the linear classical statistical models employed for
the comparison. Sun et al. (2008) applied a novel NN technique called extreme learning
machine (ELM) to investigate the relationship between sales amounts and some
Demand
forecasting
21
significant factors that affect demand in fashion retailing. In ELM, the input weights
(linking the input layer to the hidden layer) and hidden biases are randomly chosen, and
the output weights (linking the hidden layer to the output layer) are analytically
determined by using the Moore-Penrose (MP) generalised inverse. Efendigil et al. (2009)
presented a comparative forecasting methodology regarding uncertain customer
demands in a multi-level supply chain structure using neural techniques. The objective
of the paper is to propose a new forecasting mechanism that is modelled by artificial
intelligence (AI) approaches, including the comparison of both artificial NNs and ANFIS
techniques to manage the fuzzy demand with incomplete information. Their results
indicated that the ANFIS performs more effectively than does the artificial NN structure
in estimation of the more reliable forecasts for their case. Huang (2009) used Monte Carlo
simulation to solve the demand forecasting problem in the marketplace with an
expansive range of products with high random volatility of demand and correlations
between demands of product. Sayed et al. (2009) introduced a decision support system
for industrial companies using a hybrid forecast model based on combining statistical
forecasting methods and an improved genetic algorithm to model demand factors with
the demand series. They stated that the use of a combined intelligent model is necessary
for providing accurate forecasting specially for complex environments that have
different demand factors. Thomassey (2010) studied on sales forecasts in clothing
industry in terms of distributors. The paper also includes the comparison of fuzzy logic,
NNs and data mining methods.
JM2
9,1
22
