New Fashion Products Performance Forecasting: A Survey on Evolutions, Models and Emerging Trends

Andrea Avogaro, Luigi Capogrosso, Andrea Toaiari, Franco Fummi, Marco Cristani · 2025

Read New Fashion Products Performance Forecasting: A Survey on Evolutions, Models and Emerging Trends by Andrea Avogaro, Luigi Capogrosso, Andrea Toaiari, Franco Fummi, Marco Cristani (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2501-10324.txt.

Introduction

Andrea Avogaro and four co-authors survey research from 2025 on forecasting how new fashion items will sell. Fast fashion makes huge runs of new styles, which wastes cloth and harms the environment. Better forecasts could cut overproduction, but taste shifts fast and old trends return in new forms, so the task is hard.

Their survey sorts past work into the first full taxonomy of learning-based methods for this task. It reviews ways to join image, text, and sales data, and lists the data sets teams can use. The authors close with open problems and paths for greener, sharper forecasting tools. The survey follows a standard, open method for sorting and screening the past studies it covers.

Catalog record

CREATORAndrea Avogaro, Luigi Capogrosso, Andrea Toaiari, Franco Fummi, Marco Cristani
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE44 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2501-10324

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

New Fashion Products Performance Forecasting: A Survey on Evolutions, Models and Emerging Trends Andrea Avogaro, Luigi Capogrosso, Andrea Toaiari, Franco Fummi, Marco Cristani Abstract The fast fashion industry’s insatiable demand for new styles and rapid production cycles has led to a significant environmental burden. Overproduction, excessive waste, and harmful chemicals have contributed to the negative environmental impact of the industry. To mitigate these issues, a paradigm shift that prioritizes sustainability and efficiency is urgently needed. Integrating learning-based predictive analytics into the fashion industry represents a significant opportunity to address environmental challenges and drive sustainable practices. By forecasting fashion trends and optimizing production, brands can reduce their ecological footprint while remaining competitive in a rapidly changing market. However, one of the key challenges in forecasting fashion sales is the dynamic nature of consumer preferences. Fashion is acyclical, with trends constantly evolving and resurfacing. In addition, cultural changes and unexpected events can disrupt established patterns. This problem is also known as New Fashion Products Performance Forecasting (NFPPF) , and it has recently gained more and more interest in the global research landscape. Given its multidisciplinary nature, the field of NFPPF has been approached from many different angles. This comprehensive survey wishes to provide an up-to-date overview that focuses on learning-based NFPPF strategies. The survey is based on the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) methodological flow, allowing for a systematic and complete literature review. In particular, we propose the first taxonomy that covers the learning panorama for NFPPF , examining in detail the different methodologies used to increase the amount of multimodal information, as well as the state-of-the-art available datasets. Finally, we discuss the challenges and future directions. keywords New Fashion Products Performance Forecasting, Computer Vision for Fashion, Multimodal Time Series Forecasting. 1 Introduction The challenge of predicting the sales performance of new fashion products, known as the New Fashion Products Performance Forecasting (NFPPF) problem, is a critical issue in the apparel industry Cho et al. (2014) ; Beheshti-Kashi et al. (2014) ; Singh et al. (2019) , where companies must accurately forecast future sales of new products while having little or no historical data. In an industry characterized by rapidly changing consumer trends and preferences, forecasting demand for new items is essential for optimizing production, inventory management, and marketing strategies. Traditional time series forecasting methods, such as ARIMA Box et al. (2015) or other statistical models, such as Hyndman et al. (2008) ; Box et al. (2015) ; Taylor and Letham (2018) , heavily depend on historical data. Similarly, newer Deep Learning (DL) models Salinas et al. (2020) ; Oreshkin et al. (2020) operate under the same assumption, which may prove inadequate when forecasting new products without prior sales history. This creates a unique challenge with far-reaching implications for supply chain efficiency, financial planning, and sustainability. Addressing this problem, first introduced by Ren et al. (2016) ; Arvan et al. (2019) ; Lara-Benítez et al. (2021) , is crucial due to several issues. Firstly, inaccurate demand forecasting can lead to overproduction, resulting in overstocking, financial losses, and increased waste Fisher et al. (2001) ; Taplin (2014) . This issue is particularly severe in the fashion industry, which is under increasing scrutiny due to its environmental impact due to excess production and high return rates V&A () . Secondly, underestimating demand can result in inventory shortages, missed sales opportunities, and dissatisfied customers. As a result, balancing supply and demand is essential, particularly in the face of volatile fashion trends and unpredictable consumer behavior Masyhuri (2022) . The problem of forecasting demand for new products. The NFPPF task can have several different hues, trying to cover different real-world scenarios in the fast fashion industry. In this Section, we will try to define different use cases with different motivations, which can be grouped under NFPPF . The first scenario consists of forecasting the demand for new products, in which models need to estimate future sales of a product that has never been sold. In this case, the challenge is to find patterns and similarities with older, more successful products while incorporating real-time market trends Singh et al. (2019) . When only a small number of initial observations are available, we can identify a second scenario of NFPPF Ekambaram et al. (2020) ; Skenderi et al. (2024) . Here, the challenge is to make reliable forecasts with limited data, requiring models that can learn quickly from sparse sales data and adapt to changing trends. These forecasts are critical to making timely decisions about replenishment or adapting marketing strategies after the item is released. Finally, the third use case involves forecasting a product that has not yet been released and is still in the development phase, helping companies in the decision-making process Joppi et al. (2022) . It involves building personalized models that understand the behavior, preferences, and purchase history of customers, allowing companies to suggest the most relevant products to each user. Personalized recommendations increase customer satisfaction and stimulate sales by improving the relevance of offers. This could be done to decide whether a product is worth producing or not, to propose garments only with a certain performance in the market. In all three scenarios, the need for models that can effectively predict the amount of product that will be sold is shared. However, the requirements with respect to the application case are different and satisfy different needs of clothing manufacturers. Section 3 lists several models that fall into these three specific cases also in relation to the datasets used in the experimental section. The role of learning-based models. To effectively address the challenge of forecasting the performance of new products in the fashion industry, Machine Learning (ML) and DL models offer significant advantages over traditional methods. Unlike classic time series forecasting models, which are limited in predicting new item sales with historical data, DL models can leverage a range of product characteristics such as color, style, and category, but also with other marketing information such as discounts, promotions, and release dates. In this way, they can emulate expert judgment, which has traditionally been used to predict new fashion items Hyndman and Athanasopoulos (2021) . Through this emulation, these models improve the forecasting process by capturing similarities between new and existing products on the market. 1.1 NFPPF: A Multimodal Problem Figure 1: a) A standard forecasting setup, where an evergreen item has past observations to exploit, e.g. , # sales. b) NFPPF problem, where no past observations are available, and exogenous data must be considered. The figure is from Joppi et al. (2022) . One of the major breakthroughs in this field has been the integration of visual data. Although early models focused primarily on textual attributes Bahdanau et al. (2015) ; Singh et al. (2019) ; Craparotta et al. (2019) , such as product description, adding visual information, such as product images, has significantly improved accuracy. Visual features often provide a richer and more nuanced understanding of products than textual descriptions alone, making image-based models more effective in predicting consumer demand Ekambaram et al. (2020) . This shift reflects the realization that, in the fashion industry, appearance plays a key role in determining a product’s success in the marketplace. In addition

Read the complete text