Possibilities of Sale Forecasting Textile Products with a Short Life Cycle

Abstract

Almost 115 million tons of fibers comprising almost 90 million tons of chemical fibers were produced in the world in 2021, which are mainly used for the production of clothing and footwear. A total of 30% of textile and apparel products are never sold, which means extreme waste production. This article points out possibilities of forecasting the sales of clothing in the case of one relatively large online store. Inadequate stocks of textile products in the company lead to losses and overstock leads to the need to sell products at a discount, which is undesirable and not sustainable for the company. Therefore, the aim of this research is to design a forecasting system based on classical methods (with emphasis on seasonality) and its verification in practice. The results were verified directly with the real sale or with results from a model based on a neural network. The problem with textile products is that they have a short life cycle, i.e., the length of the life cycle is approximately half a year, and a high seasonality is also presented. Therefore, the seasonal indices and Holt–Winters methods (multiplication and additional approaches) were used for forecasting products. Ultimately, this model could contribute to reducing the loss of unsold goods and thus reduce the waste of resources and increase the use of goods in other similar companies.

1. Introduction

In the beginning, it would be appropriate to mention a few facts about why this topic is interesting from the point of sustainability. Textile and apparel industries are some of the fastest-growing industries, providing employment to millions of people. Estimation according to the Ellen MacArthur Foundation [1] is that globally, the clothing industry employs more than 300 million people in all its activities within the chain, which includes people that are in design, transport, distribution, and retail companies connected to the fashion industry. There are 1.3 million people employed in the textile and clothing industry at approx. 143,000 companies in Europe [2]. It grew from USD 530.97 billion in 2021 to USD 575.06 billion in 2022 at a compound annual growth rate (CAGR) of 8.3% [3]. The market is expected to grow to USD 760.21 billion in 2026 at a CAGR of 7.2%. On the other hand, textile production uses a lot of water and land for growing cotton and other fibers. The global textile and clothing industry is estimated to have used 79 billion cubic meters of water in 2015, while the water demand of the entire EU economy was 266 billion cubic meters in 2017. It is estimated that 2700 L of fresh water is needed to produce one cotton T-shirt, which corresponds to the need for drinking water for one person for 2.5 years [4]. Greenhouse gas emissions during the production of clothing and footwear constitute another negative impact on the environment, accounting for roughly 10% of global greenhouse gas emissions.

The next topic to think about, in connection with the textile and apparel industry, is the considerable waste of resources. This is proven by the authors’ study [5] and other facts.

• Consumers in the United Kingdom have an estimated USD 46.7 billion worth of unworn clothes in their closets [6]. Another study with 18,000 members of various households in 20 countries, conducted by the relocation and removals company Movinga, revealed that the majority of consumers around the world are highly delusional about how much they own versus how much they actually wear. This study was generalized to the results, showing that people do not wear at least 50 percent of their wardrobes [7].
• Contrary to general hope, a lot of our clothes are not recyclable. It is estimated that only 13.6% of the clothes and shoes that are thrown away in the US will be further recycled, and only 12% of the material from these used clothes is recycled. A total of 12% of the material is just shredded and used as furniture stuffing, some kind of insulation, or cleaning clothes. Not even 1% is collected and used for the production of new clothes [8,9].

This trend of wasting resources is certainly not easy to stop or mitigate. However, overproduction, i.e., offer exceeding demand, is evident. This article proposes a simple solution through the analysis and prediction of the consumption of textile and apparel products in a local e-shop offering these kinds of products and is aimed at the Czech and Slovak markets. This can be inspiring for any store, anywhere in the world, that has unnecessary stocks of goods that are not in high demand, or insufficient stocks of goods that are in short supply or sell out quickly. This study analyzes some selected clothes that are considered common and key anywhere in the world.

Of course, this way of reducing sources is not the only one because this fact of wasting resources in the textile industry has already been noticed by many authors in their publications, and thus they contribute with their research not only to reduce the consumption of new resources but especially to increase the share of material recycling [10]. For example, Chaka [11] pointed out that the conversion of spinning waste into valuable products will speed up the growth and development by creating additional profit and also contributing to the environment. Baruque-Ramos et al. also pointed to the reduction in resource consumption, mainly to increase public opinion about the need for saving by approaching responsible lifestyles and environmental awareness and the refusal to waste resources in general. They do not even stop thinking about the production itself, where the main socioenvironmental benefits are related to the training of labor and local income generation [12]. Stone et al. focused on comparing the negative impacts on freshwater ecosystems from the way we produce, use, and dispose of textiles. A comparison was made from the point of view of the risks associated with the use of natural or synthetic fibers. Woolen (natural) textiles pose the most risk during the production phase, while PET (synthetic) textiles pose the most risk during the use and disposal phases [13]. Similar research for marine ecosystems has been proposed by Agamuthu, P. et al. They noted that tackling the marine debris crisis is not a straightforward, one-size-fits-all solution, but rather an integrated and continuous effort required at local, regional, and global levels [14]. Provin, A.P. et al. have also dealt with the idea of reducing raw material inputs in the textile industry by the use of alternative biomaterials in their research. The bacteria of the Komagataeibacter xylinus family, which is present in the probiotic drink Kombucha^®, is a great source for the production of bacterial cellulose (BC) and has the potential to replace current fabrics in the production of clothing and accessories [15].

In this industry, it is not only about raw material wast, but also about a continuous increase in energy consumption and the Energy Intensity (EI) index, primarily electrical and thermal energy, which was noticed by Jaitiang, T. et al. They studied four cases of energy sustainability and the implementation of renewable energy in the sector: (1) Business-As-Usual (BAU); (2) switching to higher-efficiency equipment and waste heat recovery; (3) solar PV (PhotoVoltaic) utilization; and (4) the combination of the second and third case. They recommend implementing more renewable energy actions concerning the textile industry or modifying policy and technology [16]. The case of water pollution by the textile industry (also mentioned above) has been suggested by Behera, et al., where they made an extensive overview of many specific technologies with the aim to initiate influential policies among the research communities to fight against the widespread risk of unwilling treatment practices of toxic organic pollutants generated from textile industrial companies. However, water pollution does not end only with the production itself, it also uses products made from synthetic fabrics. Micro-plastic fibers (MPFs) can also get into wastewater by washing clothes in households [17,18,19]. Cai, Y.P. et al. claimed that there is a threat of MPFs, and the aim of their study was to investigate the presence of MPF in various semi-finished and finished polyester textile products. On average, five times more MPFs could be obtained from textiles with treated surfaces (such as fleece) than from textiles with untreated surfaces. This means that abrasive friction during manufacturing can also create MPFs. For example, cutting a textile with scissors can generate more extracted MPFs than laser-cut textiles. This made it possible to quantitatively distinguish the MPF contributions from the textile surface versus those originating from the edges of the textiles [20]. Also, Ramasamy, R. and Subramanian, R.B. looked for solutions to reduce the formation of MPFs in their review. They have found that the use of finer-count yarns with filaments and compact structures reduces microfiber shedding. A significant increase is noted in the microfiber reduction percentage after the chemical (coating) finishing process [21]. This issue of MPF threats motivated Henry, B. et al. to introduce a critical review of factors affecting the release of microfibers from fabrics and the risks on ecological systems, potentially human health [22]. Araque-Gonzalez, G. et al. proposed a modern approach to reduce input requirements, especially in the clothing industry for small- and medium-sized enterprises. They propose an industrial production model with a focus on Industry 4.0 (big data and decision-making analysis) that allows for improving procedures, jobs, and related costs within the study organization [23].

The worldwide expansion of the textile and apparel industries also results in the development of related industries, such as transport. It has a secondary impact on the environment, and Dhonde, B. and Patel, C. have found that well-organized trip planning and optimized utilization of payload capacities can reduce vehicular emission generated from commercial good movements in the textile industry to 2/3 of its current levels [24].

Forecasting in the textile industry is crucial because it helps businesses stay ahead of the competition and plan future growth. Companies can ensure that they have the right inventory in place to meet customer needs and avoid stock shortages or overproduction by predicting demand. This is particularly important in the fast-paced and ever-changing world of fashion where trends and styles can change quickly [25].

In addition to inventory management, forecasting can also help textile companies make decisions about new product development, market expansion, and production processes. By anticipating demand, businesses can determine which products will be most popular and allocate resources accordingly. They can also identify new markets and opportunities for growth, helping to drive long-term success. Jiang, S. and Liu, Y. forecasted the variable demand for textile products for small and medium enterprises in China. According to them, the use of gray models of forecasting appears to be a very good choice [26]. Thomassey, S. et al. solved forecasting to improve supply chain management for textile companies in uncertain environments by use of simulation [27]. They also presented a forecasting model that consists of several methods and performs forecasts for different horizons at different levels of sales aggregation. This system is based on soft computing techniques, such as fuzzy logic, neural networks, and evolutionary procedures, enabling the processing of uncertain data [28]. Research does not have to include only the use of artificial intelligence methods because Sabir, E.C. and Batuk, E. verified the creation of forecasts using classic methods and found that the Basic Exponential Smoothing method is an unsuitable demand forecasting method, while the Trend Corrected Exponential Smoothing method Winter’s method might be used for a demand forecasting system in textile dyeing–finishing mills [29]. Yuan, B.J.C. and Chang, P.C. published their study of forecasting using the Delphi method to present results for the industry, government, and academics in their continuous promotion and research of the textile industry [30].

As described above, forecasting in any form is also applied to the textile industry for the purpose of the economic use of resources to support sustainability across various sectors, such as:

• Energy and raw material consumption. Accurate forecasting helps organizations predict their energy and material needs. Businesses can reduce waste and optimize their use of resources by planning their consumption efficiently.
• Supply chain management. Planning sustainable supply chains involves optimizing transportation routes, reducing emissions, and choosing environmentally friendly suppliers.
• Financial efficiency. Sustainable practices often lead to cost savings in the long term. Planning these initiatives can help organizations make informed decisions that balance environmental and financial objectives.
• Regulatory compliance. Forecasting allows organizations to anticipate and adapt to changing regulations related to sustainability. This proactive approach helps avoid non-compliance issues and potential fines.
• Technology adoption. Organizations can plan the adoption of sustainable technologies and innovations, such as energy-efficient equipment or eco-friendly manufacturing processes, contributing to long-term sustainability goals.
• Resilience planning. Forecasting can assist in anticipating the impact of climate change on operations and supply chains. Planning resilience against climate-related risks is a key aspect of sustainable business practices.

The fact that the headquarters of the Czech and Slovak e-shop company needed to implement a certain planning system inspired the authors to apply forecasting methods to determine the approximate needs of the selected items of the assortment within the annual time horizon. This should contribute to savings from the side of storage (e.g., unnecessary overstocks or deficiencies, which can ultimately result in customer losses). Furthermore, from the side of the transport of goods, there is also a space for reducing costs, such as reducing returns, i.e., excess goods or urgent shipments. In the global world of the textile industry, it is difficult to enforce environmental aspects and requirements to reduce the use of resources, but as practice has shown, some companies still work without conceptual planning with regard to sustainability. Implementing forecasting into the planning system of the e-shop company that sells clothing, shoes, and accessories seemed to be a good idea for the above-mentioned reason. The goal of this research was to design a forecasting system based on classical methods (with an emphasis on seasonality) and verify it in practice. In order to increase the objectivity of the results, a combined model was used to integrate the results of three forecasting methods. The following chapters describe the principle of the model and the verification of the results directly with reality or with results from a model based on a neural network.

2. Materials and Methods

In today’s data-driven business environment, quantitative methods are still widely used to make accurate and reliable forecasts. These methods use statistical and mathematical models to analyze historical data and make predictions about future trends and patterns. Quantitative methods for forecasting in the textile industry are based on regression analysis and time series analysis. This article includes methods such as the Holt–Winters (multiplicative and additive approach) method [31,32] and seasonal indices. The Holt–Winter smoothing parameters (α, β, γ) were set according to the minimum RMSE. These methods are particularly useful for industries with clear patterns in demand and seasonality, which is typical in the textile and apparel industry. Textile companies can make informed predictions about future demand for their products and make strategic decisions to improve their bottom line by analyzing historical sales data, production data, and economic indicators [33,34].

As was described above, the Holt–Winters forecasting method and the seasonal indices method can be applied in the textile industry for demand forecasting and inventory management. Demand for products is influenced by various factors such as seasonality, fashion trends, and economic conditions in the textile industry. These methods can be used to forecast demand for specific textile products, such as cotton shirts or woolen sweaters, taking into account seasonal patterns and trends in consumer behavior. This information can be used to adjust production and inventory levels, reducing the risk of overstocking or stock shortages. Additionally, these methods can be used to forecast raw material demand, allowing textile companies to plan and manage their supply chain more effectively.

The uniqueness of this forecasting proposal lies in the fact that the results from individual forecasting methods are combined into one common result [35]. A weighted average was used with a variable weight, so that the more accurate result of the individual method is (compared to the actual values), the weight of such forecast is higher. Since the overall forecast consists of a combination of the results of three forecasting methods (Holt–Winters, two approaches, and seasonal indices), their results can be generally written as:

-

Holt–Winters (multiplicative approach) …. Forecast1 (F1);

-

Holt–Winters (additive approach) ………… Forecast2 (F2);

-

Seasonal indices……………………………… Forecast3 (F3).

Weights for individual forecasts are calculated according to the accuracy of particular forecasts (Equation (1)). Let us take into account the values of the accuracy of the MAPE. Then:

$$w_{Fi}=\frac{1-\left(\frac{{MAPE}_{Fi}}{{MAPE}_{F1}+{MAPE}_{F2}+{MAPE}_{F3}}\right)}{2}$$

(1)

where

w_Fi—the weight of the particular (i) forecast method;

MAPE_Fi—the accuracy of the particular (i) forecast method.

In this way, a higher accuracy of the overall forecast is achieved for all products. The total forecast is calculated by Equation (2):

$$TF={\sum }_{i=1}^{3}Fi {\times w}_{Fi}$$

(2)

where

TF—total forecast.

The objects of the forecast were chosen in the form of a group of products that have a common meaning, approximate appearance, and functionality and thus can be classified into a common category. Therefore, specific items were not selected because the behavior of selling particular items is very changeable or volatile [36]. The product groups were chosen to cover the entire seasonal life cycle of clothing. In the winter period, sales of coats were significant; in spring and summer, dresses, short-sleeved T-shirts, sneakers, and jeans were sold well, and leather jackets came to the forefront of sales at the turn of autumn. It can be concluded that the life cycle of the mentioned clothing also extends into the next season but often with a lower sales volume. The products are chosen so that their life cycle is the longest and at the same time they are among the best-selling products. So, the life cycle of these products can be characterized as two seasons. The selection of the products was performed by the brainstorming of the sales manager, two sellers working in the e-shop, and one main supplier. The main criterion was the volume of sales, i.e., “best-selling” items, which also means a relatively high turnover. Another secondary criterion was substitutability, i.e., whether the particular product has a replacement in the form of a new product with similar properties to the previous one. The data for calculating the forecast come directly from the company operating the e-shop. Since this company operates its e-shop in the Czech Republic as well as in Slovakia, two total forecasts will be created for each group of products: the forecast “CZ + SK” (this means for both markets together) and separately “SK” (for the Slovak market only).

The data that were selected for the forecast were from the period between 2017 and 2018, and the forecast was made for 2019. Although these data are current, they come from a relatively stable period with regard to the view of this market segment. The later data were influenced by the crisis caused by COVID-19, and for that reason, it would not be appropriate to use these data. Forecasts on the basis mentioned above would not be relevant and applicable. The period 2022 to the present (i.e., the post-COVID-19 crisis period) seems to resemble 2018–2019, but now there is a lack of insufficient data.

3. Objects and Results

The groups of products that were selected for the analysis and solution of the forecast represent a wide range of assortments offered by the company engaged in the sale of textiles and apparel products.

• The first group of products: Coats_1. Black coats were chosen because they were the best-selling items in a given period, i.e., from January to June, when the end of their life cycle is also expected. The life cycle lasted 6 months in the 2017 season and only 5 months in the 2018 season. Taking into account the interseasonal period, the life cycle lasted a total of 26 weeks. The chosen coats were from Vero Moda, which is a Danish fashion brand for women.
• The second group of products: Coats_2. The sale of winter coats was from January to June in 2017 and from January to May in 2018. The total sale lasts 23 weeks and starts in the first week of the year. The product group consists of Dorothy Perkins brand coats. This brand is very popular and is characterized by its high quality. It is one of the most successful British fashion brands for women, with more than 100 years of tradition.
• The third group of products: Dresses in blue with a floral pattern. The sale in 2017 was from May to November and in 2018 from May to December. The sale started on the 19th week and ended on the 50th week in the given years. The products are Closet brands. This brand has been on the market since 1996 and is a well-known British brand that is especially popular among celebrities.
• The fourth group of products: White sneakers. In 2017, they were sold from February to August, and in 2018, from January to August. The total life cycle lasts 31 weeks. It starts on the 5th week and ends on the 35th week. The common feature of sneakers is their simple cut. They are from the Swedish brand Vagabond, which was first for men and later also for women. They are characterized by high quality and a wide range of styles. The company was founded in 1968. In addition to footwear, it also offers leather accessories in more than 40 countries around the world.
• The fifth group of products: Black T-shirts for men. They were sold from January to November. This is the longest life cycle of the selected products. It lasts 44 weeks and starts on the 2nd week and ends on the 45th week in the given years. This product is the Jack & Jones brand. It is a Danish brand that makes models for men. It was founded in 1990 and currently has more than 1000 shops in 38 countries and cooperates with more than 1000 wholesalers.
• The sixth class of products: Red T-shirts for women with short sleeves. They are Vero Moda brands. Sales in 2017 were from March to September and in 2018 from February to September. The total life cycle lasted 33 weeks. It started on the 6th week and ended on the 38th week.
• The seventh group of products: Brown leather jackets. It is a universal leather jacket and is very popular among women. The zipper is located in the middle and the collar forms a waist with buttons. The sales in 2017 and 2018 were from February to September and lasted a total of 32 weeks. They started on week 9 and ended on week 40 in those years. The jackets are from Only brand, which is Danish fashion for women. It was founded in 1995. It belongs to the Bestseller Company, which also includes the brands Vero Moda and VILA, which are mentioned among the products.
• The eighth group of products: White shirts. They are classic white button-down shirts. Sales in 2017 were from April to November and in 2018 from March to November. Missing data were filled in with the number one. The total life cycle lasted 34 weeks. It started on the 12th week and ended on the 45th week. The products are from the Dorothy Perkins brand, which is characterized by a second class of products.
• The ninth group of products: Dark blue jackets with a classic one-button cut. The sleeves are gently layered. The sale was in 2017 and 2018 from January to June. The total life cycle lasted 24 weeks. It started on week 1 and ended on week 24 in the given years. Products are from the Vero Moda brand.
• The tenth group of products: Simple skinny jeans with a torn effect. Jeans of this cut have been in fashion for several seasons. The sale took place in 2017 from May to October and in 2018 from April to November. The total life cycle lasted 29 weeks. It started on the 17th and ended on the 45th week.

Table 1. Registered periods of the product’s sale.

Product	Date of Registered Sale (Period of Sale)
	First Period (2017)	Second Period (2018)
Product 1: Coats_1	1 January–30 June	1 January–30 June
Product 2: Coats_2	1 January–31 May	1 January–3 June
Product 3: Dresses in blue	7 May–30 November	8 May–5 December
Product 4: White sneakers	1 February–31 August	1 February–31 August
Product 5: Black T-shirt	10 January–31 October	11 January–31 October
Product 6: Red T-shirt	1 Febuary–15 September	1 February–16 September
Product 7: Brown leather jacket	1 March–30 September	1 March–30 September
Product 8: White shirts	15 March–31 October	15 March–31 October
Product 9: Dark blue jacket	1st January–17 June	1 January–16 June
Product 10: Simple skinny jeans	1st May–31 October	1 May–31 October

shows the sale periods of individual products in their respective weeks in graphic reports.

Table 2. Transformation of periods to weeks in graph diagrams.

Product	X-axis (Weeks in Graph Diagrams)	Forecast (Weeks in Diagram)
Product 1: Coats_1	1–26 and 27–52	53–78
Product 2: Coats_2	1–23 and 24–46	47–69
Product 3: Dresses in blue	1–32 and 33–64	65–96
Product 4: White sneakers	1–31 and 32–62	63–93
Product 5: Black T-shirt	1–44 and 45–88	89–132
Product 6: Red T-shirt	1–33 and 34–66	67–99
Product 7: Brown leather jacket	1–32 and 33–64	65–96
Product 8: White shirts	1–34 and 35–68	69–102
Product 9: Dark blue jacket	1–24 and 25–48	49–72
Product 10: Simple skinny jeans	1–28 and 29–56	57–84

shows the transformation of periods to weeks in graph diagrams (Figure 1).

3.1. Product Characteristics and Forecast

Coats_1—The popularity of black coats varies depending on the season. Generally, coats are more popular during colder months in the SK and CZ regions.

Coats_2—there is a big similarity to Coats_1. They have the same features of the sale behavior compared with the previous product, Coats_1. Having multiple coats (winter coats) to choose from can be useful for different weather conditions or outfit styles.

Dresses in blue—Blue is a popular color, and dresses are versatile clothing items that can be worn for various occasions, so dresses in blue can be relatively popular.

White sneakers—White sneakers have been a popular footwear trend in recent years, and they continue to be a popular choice for their versatility and simplicity.

Black T-shirt—Black is a classic and versatile color, and T-shirts are a staple item in most people’s wardrobes.

Red T-shirt—Red is a bold and eye-catching color, and T-shirts are a casual and comfortable item. Red T-shirts can be popular depending on the fashion trend and personal preferences.

Brown leather jacket—Leather jackets have been a timeless fashion staple for many years, and brown is versatile and classic and can add sophistication to any outfit.

White shirts—White shirts are classic wear that can be dressed up for serious or casual occasions, and this makes them a popular clothing item in most wardrobes.

Dark blue jacket—Dark blue is also a versatile color that can pair well with many different outfits, and jackets are essential outerwear for colder weather.

Simple skinny jeans—Skinny jeans have been a popular clothing item for several years, and they continue to be a staple item in most wardrobes for younger and older people. Simple designs can be versatile and easily paired with other clothing items, making skinny jeans a best-loved choice.

Figure 1 comprises all diagrams of product sales as introduced above. It displays the total (combined) forecast separately for the SK and SK + CZ regions.

In Figure 2, there are graphic diagrams showing the forecast for 2019 and the actual sales in 2019 for the SK and SK + CZ markets in more detail. The differences between actual sales and the forecast are the inputs to the calculation of the forecast errors, which are quantified below.

3.2. Calculation of Forecasting Error and Its Relevancy

The quantification of the forecasting errors is calculated based on a comparison with the actual sale values in 2019 and the forecasts that were calculated for the year 2019, which are mentioned in the previous subsection. Forecasting errors are calculated using generally known indicators, i.e., the Mean Absolute Percentage Error (MAPE) and the Root-Mean-Squared Error (RMSE). They are important metrics for evaluating the accuracy of forecasting models. Due to the fact that these metrics are the most commonly used indicators, this article does not include the exact procedures and formulas for their calculation.

Table 3. The RMSE and MAPE of the products in both market areas.

Product	RMSE		MAPE
	SK	SK + CZ	SK	SK + CZ
Product 1: Coats_1	1.019	2.236	19.04%	31.24%
Product 2: Coats_2	0.834	3.369	25.51%	35.49%
Product 3: Dresses in blue	1.668	3.841	29.71%	23.49%
Product 4: White sneakers	1.136	2.806	28.74%	22.59%
Product 5: Black T-shirt	1.638	2.900	29.31%	12.76%
Product 6: Red T-shirt	1.155	2.752	27.03%	24.26%
Product 7: Brown leather jacket	1.250	2.710	28.40%	12.89%
Product 8: White shirts	1.361	2.532	27.31%	14.18%
Product 9: Dark blue jacket	1.323	3.123	26.63%	17.43%
Product 10: Simple skinny jeans	1.336	3.808	24.72%	20.64%

shows the RMSE and MAPE of the mentioned products in both market areas.

4. Discussion

The sales analysis and prediction of the first product, black coats (Coats_1), indicated that there would be a decrease in sales. There is a year-on-year decrease of more than a third (from 332 units in 2017 to 204 units in 2018) in the SK + CZ markets. The forecast showed that sales in the next year will drop slightly to 141 units. The reality was not far from the forecast, as shown by the actual sales of 147 pieces of clothing. The situation was similar in the single SK market, but the decline was not so striking. The sale of 37 pieces of clothing in 2017 decreased to 25 pieces in 2018. According to the forecast, a slight decrease to 22 pieces was expected, but the actual sale was slightly higher, up to 29 pieces. The overall sales outlook for this specific product next year is expected to be replaced by redesigned products, which may lead to a more modest increase in sales.

In the analysis and forecasting of the second product of winter coats (Coats_2), a slight year-on-year increase in sales in the SK + CZ markets was recorded (240 units were sold in 2017 and 252 units were sold in 2018), and the peak sales were during the first week of January 2018. The forecast points to a slight increase, but the number of units sold in 2019 was almost the same as in 2018. A similar situation was also recorded in the SK market, where sales were at the level of 34 units compared to the forecast of 32 units. Although it was the same model from 2018, sales in 2019 were supported by colder weather in January 2019 than in 2018 [37].

The analysis of the third product (dresses in blue) indicated the typical behavior of a short-term fashion hit, where peak sales were registered at the beginning of June, which is considered the first summer month in SK and CZ. However, the forecast for this product for 2019 indicates a decrease, which actually happened, and sales in 2019 were 208 units, which is almost half of the sales in 2018. This phenomenon can be seen in the gradual saturation of the market despite the frequent change in the cut of this dress. A similar situation can also be seen in the SK market.

The fourth product (white sneakers) is traditional, unisex, and popular for all generations, and after analysis, it can be seen that the time series is typically seasonal, and sale volumes in individual years are relatively the same—at the level of 450–550 pcs. The sales forecast for 2019 for the SK + CZ market and the single SK market was slightly optimistic compared to actual sales, with differences of thirty-two units in the SK + CZ market and eight units in the SK market, which is a 7.3% difference in the SK + CZ market and 11.0% in the SK market.

Another traditional clothing item for men is the black T-shirt, which also shows signs of seasonality with top sales in the second half of May. It is a product that has one of the longest life cycles in the clothing industry because it is redesigned very little and, therefore, it is possible to claim that it is the same product during the three years of the sale observation. The analysis shows that the volume of sales increases slightly from year to year, and the assumption resulting from the forecast points to an increase in sales in 2019. So, the sales in the SK + CZ market were 890 units in 2017, 1020 units in 2018, and 1024 units were sold in 2019. A total of 107 units were sold on the SK market in 2017, 155 units in 2018, and 180 units in 2019.

The sixth analysis and forecast were created for another similar and popular product (red T-shirt). This is a relatively stable situation compared to 2017 and 2018. In the SK + CZ market and SK market, there was a slight increase in sales in 2018. The forecast and reality in both cases pointed to the stability of the sales of this product, but still, a slight decrease in sales for 2019 was registered by 6.4% in the SK + CZ market and by 4.2% in the SK market.

For the seventh product, there was an interesting double season of interest during the year. The first minor seasonal interest occurred in the period of changeable weather at the beginning of April, and even more interesting is the fact that in the middle of the summer, there was a significant seasonal interest; that is, when the month changed from July to August. It is possible that these were sales with discounted prices, but this information is not known retrospectively. In the SK + CZ market, the forecast also showed that a slight increase in sales continued (515 pcs in 2017, 536 pcs in 2018, and 570 in 2019). In the SK market, there was a relatively significant drop in sales in 2019 by 28 pcs, which is a decrease of almost 30%. This may be affected by the extremely warm weather at the end of July 2019 [38]. Otherwise, thanks to the changes in the cut of these jackets, there is constant interest in sales.

The eighth product (white shirts) is also a product in which there is constant high interest because it is a part of formal clothing. A constant slight increase in sales was also expected, which was confirmed by the forecast and the sales in 2019 in the SK + CZ market in the SK market. The year-on-year increase in the SK + CZ market is 24% on average, and the year-on-year increase in sales in the SK market is 20.6% on average.

The analysis and sales forecast of the ninth product (dark blue jacket) did not indicate any significant changes. A slight decrease in sales was predicted for the SK + CZ market, which was also indicated by the change from 381 pcs in 2017 to 328 pcs in 2019. However, the forecast was a bit more pessimistic, but in 2019, in the second half of March, there was short-term colder weather in several places in Czechia [39]. On the contrary, an increase was recorded in the SK market between 2017 and 2018 and a slight decrease in sales in 2019 of 13.5%.

A similar situation compared to the ninth product is also found in the last tenth item (simple skinny jeans). The forecast was more pessimistic than the sales themselves, which peaked in the SK + CZ market in the second half of August and exceeded the expected sales in this period. Overall, however, a decrease in sales was recorded compared to 2018 by 132 units (25.6%). A decrease in sales was also expected in the SK market, but in reality, it was smaller than expected compared to 2018 by 16 units (17%).

However, the authors must state that the given forecasts for individual products are limited due to the amount of data that were available. Nevertheless, for reasons of verification, evaluation, and relevance of the forecasts, a “mirror” forecast was also created for product 1 for the SK and SK + CZ markets. This forecast was created through an NAR (Nonlinear Auto-Regressive) model as a type of neural network architecture designed for time series prediction and forecasting using MATLAB. NAR models can be quite effective for capturing complex temporal patterns in time series data. The FitNet (Function-Fitting Network) architecture of the NN was used for the calculation. FitNet is a specific type of neural network architecture designed to facilitate training and transfer learning in deep neural networks, particularly for image classification tasks [40]. These forecasts are in Figure 3: “FitNets: Hints for Thin Deep Nets” by Romero et al.

The results of the MAPE and RMSE of such forecasts are listed in

Table 4. The results of the MAPE and RMSE forecast errors.

Product	RMSE		MAPE
	SK	SK + CZ	SK	SK + CZ
Product 1: Coats_1	1.074	5.446	25.06%	40.06%

. As can be seen in the table, the accuracy of the forecast is not as high as in the case of the classic methods. This proves that there is not enough data to reliably train the model.

It is stated in some studies that a MAPE of a forecast between 10 and 20% means a good forecast. Even if the MAPE is larger in some cases for the proposed forecast, this does not necessarily mean that the forecast is not applicable. It still means a normal state up to 50% [41]. Of course, there is still a space for increasing accuracy, which the authors will continue to do.

5. Conclusions

This article is focused on the forecasting of textile products with a predominantly short life cycle. The problem consisted of an incorrect estimation of product sales, so the e-shop often had to sell off stock at a bigger discount. This was ultimately reflected in the company’s losses. The aim of the cooperation with the mentioned e-shop was to find products that would be substituted by products with similar properties in the following periods (redesigned products, products with a small change in cut, etc.). In this way, at least 10 products were selected, for which a more accurate sales plan for the following season was prepared thanks to forecasting.

The forecast was made for the year 2019 and was based on data from their sales during the years 2017 and 2018. The data were obtained from the company’s database in a weekly cycle, and the provided data were from single sales in Slovakia (SK) and from sales in the Czech Rep. and Slovakia together (SK + CZ). The combination of the Slovak and Czech markets was chosen because sales in Slovakia were not high enough to reflect significant changes. Even if the data did not cover the period for the whole year, the period when the sales of the given products were recorded was decisive. This reflected the life cycle of the selected products. The shortest life cycle of the product group was 23 weeks, and the longest life cycle was 45 weeks. The forecast was calculated using simple methods (the Holt–Winters method, both approaches, and the method of seasonal indices), which are intended for seasonal data.

It whichs also necessary to emphasize the principle of the combined forecast, which works on the principle of merging the results of several methods into one result using a weighted average with variable weights. They are calculated according to the accuracy of individual methods in the past period. It showed that the results are on average 2.5% more accurate than in the case of the fixed determination of weights (e.g., w_Fi = 1/3).

In the next work, we would like to devote expanding the product range, updating current data (analyzing sales during the global COVID-19 pandemic), and putting forecasting through the neural network methodology because, as was already mentioned above, the sales themselves are influenced of many other factors. One such important factor is the influence of the weather.

The COVID-19 period is very specific, and we assume that it will not indicate trend prospects for other years. Due to the restrictions that limited the movement of residents and the fact that many shops were closed, it was expected that the online store would have a rush of orders. The company confirmed that it had to refuse many orders due to a shortage of goods because it was unable to respond to such demand. The problem was also on the part of the suppliers because they were also unable to respond flexibly to the increased interest due to restrictions outside the territory of Slovakia and the Czech Republic. Anyway, we will try to analyze this period and evaluate possible forecasts from this period.