Supply Chain Management (SCM) involves a systematic coordination of multiple processes and activities which are crucial for the success of delivering products and services to end-users. It incorporates activities like Requirements Planning, Procurement, Manufacturing, Logistics and Delivery, Returning, Enabling, and facilitating services like finance and IT (Vairagade et al. 2019).
The function of the Supply Chain Management (SCM) is essential in the field of manufacturing, distribution, and trading in managing the balance between demand and supply. This process serves to minimize excess inventory while improving overall profitability. Thus, it contributes to further optimizing the inventory levels and improving the overall level of profitability. As accurate demand forecasting becomes a part of the supply chain management practice, effective supply chain operations and better customer satisfaction prove to be the outcomes (Kochak and Sharma 2015). This method helps organizations to keep inventory at optimal level, provide better customer service, and take well informed decisions.
The SCM demand forecasting is made through the studying of previous data and analyzing it to eventually predict the customers' future needs. This forecasting assists in decision-making and thus the inventory is maintained to prevent the waste of resources. The efficiency is enhanced in terms of cost reduction, stock value optimization and thus ultimately maximize revenues and customer satisfaction level in the supply chain management cycle (Terrada, Khaili and Ouajji 2022).
The role of Machine Learning (ML) and Deep Learning (DL) approaches cannot be overstated when it comes to demand-forecasting in the domain of supply chain management. Those approaches let the organizations use historical data and trends in order to generate accurate predictions. Machine learning (ML) employs some of the learning methods namely supervised, unsupervised and reinforcement for data analysis aimed at optimizing inventory levels, reducing waste and upscaling customer satisfaction. Deep learning (DL), the branch of machine learning (ML), brings more accuracy than traditional methods that are based on Artificial Neural Networks (ANN) and Convolutional Neural Networks (CNN) (Husna, Amin and Shah 2021).
The apparel industry provides a wide range of prospects in the field of demand forecasting via using advanced technologies like machine learning and deep learning. The introduction of historical sales info helps to improve the forecasting accuracy by integrating neural networks (Giri and Chen 2022). Since the apparel industry, especially the fast fashion part of it, shows high complexity, there is a lack of intelligent demand prediction systems (Brahmadeep and Thomassey 2016). The introduction of uncertain customer demand, unpredictable product lifespan and a variety of SKUs has the potential of being met with advance forecasting techniques that aim at improving inventory management and enhancing the supply chain efficiency.
In addition to this, Artificial intelligence (AI) allows organizations to improve prediction accuracy and reduce error rates by combining factors such as product details, color details, seasonal information, gender, and special days into forecasting models (Güven and Şimşir 2020). The integration of these cutting-edge tools allows us to optimize inventory management, minimize stockouts, and enhance customer satisfaction.
The apparel industry often deals with specific challenges of demand forecasting, which comes from factors such as short product life cycles, and unstable market trends. Implementation of machine learning and deep learning strategies, like neural networks, may help in advancing the precision of prediction through the adoption of past sales data (Giri and Chen 2022). The fast fashion concept is forced to use increased forecasting skills that consider shorter lead periods, high variability of demand, and limited access to historical data (Brahmadeep and Thomassey 2016). The bullwhip effect further escalates these challenges, more or less intensifying the consequences of too many stocks or shortages in the supply chain (Terrada, Khaili and Ouajji 2022). However, in order to survive these challenges, advanced forecasting techniques need to be applied which can handle complex multivariable problems and produce accurate predictions in short time frames.
Different techniques have been developed in time series forecasting, from classical statistical techniques to deep learning models, and each of them has its own strengths that they are best used for. For example, Autoregressive Integrated Moving Average (ARIMA) and Exponential Smoothing (ETS and Holt-Winters) are traditional statistical approaches which are widely known for their simplicity and effectiveness in capturing linear patterns and seasonality. Time series forecasting using ARIMA is a fundamental method, which is widely applied in many fields of science, including short term and long-term predictions. ARIMA's adaptability to different forecasting periods reveals its effectiveness in precise forecasts (Bektemyssova et al. 2022).
Alternatively, more advanced techniques like LSTM networks are chosen to help in dealing with complex time series prediction tasks. Necessity of this method consists of its abilities to capture long-term trends, seasonal fluctuations and random noise which further lead to accurate analysis and forecasting of historical trends in data (Kramar and Alchakov 2023).
Additionally, combining statistical techniques like ARIMA with machine learning models such as ANNs introduces a strong method in the field of time series forecasting. Through this strategy, the models employ the individual strengths of both models, consequently allowing the capture of linear and nonlinear patterns that are important. This combined technique gives more confident forecasts because it overcomes the constraints of each prediction technique (Zhang et al. 2022).
The past ten years have had a substantial effect in demand forecasting which has resulted in many research ventures concentrating on the production of accurate forecasting models. Demand forecasting is a key factor in a variety of areas such as retail, tourism, finance, supply chain management, manufacturing, and healthcare among others.
(Giri and Chen 2022) introduced an innovative demand forecasting system by adapting for the retail fashion apparel industry. The system addresses several problems, among them the short life cycle of products, volatile nature of the market demand. The developed model yields the forecast of sales and the results obtained are promising and a possible approach in the retail sector to solve the problem of forecasting.
(Dou et al. 2021) put forward an LSTM method of forecasting manufacturing industry demand, which was mainly aimed at resolving a variety of environmental factors’ complexities and diversities. Results show that deep learning turns out to be better than the traditional methods, and this thereby provides strategic developmental strategies for the manufacturing industry.
(Vairagade et al. 2019) conducted a study which compares a Random Forest model and a Keras Neural Network regression model for demand forecasting in Supply Chain Management based on grocery sales data from a South American chain store. Random Forest scored better than Neural Network showing its proficiency in forecasting sales demand and making it easy and accurate.
The traditional methods of demand forecasting are the root for the estimation of future demand amongst different industries like textiles. Apprehending the approaches, techniques, and weaknesses of these actors is an indispensable step to successful supply chain management and business decision making. The commonest statistical forecasting techniques include moving average, regression analysis, exponential smoothing, and the holt-winters forecasting.
(Fattah et al. 2018) suggested using ARIMA (1, 0, 1) model in food manufacturing where they used historical data to identify patterns and forecast demand by applying Box-Jenkins time series techniques.
The future sales are typically forecasted in the traditional demand forecasting methods using the exponential smoothing. (Fumi et al. 2013) presented an approach with which exponential decreases to past observations are assigned using a smoothing parameter (α).
The Traditional Approaches to Demand Forecasting are often based on some of the methods like Multiple Linear Regression Analysis. (Maaouane et al. 2021) built an Industrial Energy Demand model that depends on Multiple Linear Regression (MLR). This model implicates macroeconomic factors such as the rate of economic growth to predict the industrial sector's energy demand. The method is a bottom-up estimation approach for overall energy demand derived from applying data from consumed goods quantities across sub-sectors to international sources like the IEA.
Recent forecasting models have the potential to improve prediction accuracy and adaptability to market fluctuations in the dynamically changing apparel industry. Exploring these sophisticated approaches could considerably contribute to an improvement of demand forecasting techniques and a betterment of supply chain management.
The study conducted by (Vairagade et al. 2019) introduced two key models: Random Forest and Keras Neural Network. The objective of this research is to optimize demand forecasting in the area of Supply Chain Management (SCM) using machine learning methods. By meticulous evaluation employing performance metrics such as R2 score, Mean Squared Error, and Mean Absolute Error, it is shown that Random Forest has a superior performance compared to the Neural Network method. The results of this study promote the efficiency of SCM as it predicts the accurate demand of the product.
(Chandraul and Barode 2018) suggested an advanced demand prediction approach using artificial neural networks (ANN) in supply chain management systems. The model applies the Bayesian Backpropagation Training Function in conjunction with the Tansig Transfer Function to forecast the demand for a footwear distribution company. This ANN-based approach is more precise than other conventional forecasting methods when applied to short time horizons.
(Zhao 2023) introduced a novel technique for improving demand forecasting for supply chains in the realm of supply chain management. The elaborated strategy named WOA-ELM consists in the combination of the Whale Optimization Algorithm (WOA) with the Extreme Learning Machine (ELM) architecture. The combined WOA-ELM model outperforms the individual ELM model and maintains energy demand variation at the desired level.
(Watanabe, Yoshibayashi and Imaoka 2019) presented Fujitsu’s dynamic ensemble forecasting technology that applies the combination of traditional time series methods and machine learning to improve the demand forecasting accuracy and adaptability. Fujitsu further investigates the demand forecast for new products by application of text mining and machine learning techniques to its decomposition technique.
(Ren, Chan and Ram 2017) involved in a case study to examine the strategic use of both big data and qualitative insights by "Fashion Trends Inc.," a very popular fashion store in China, towards demand forecasting and inventory planning activities within the fashion industry. This technique ensures that forecasting is accurate, able to respond to market changes, and inventory levels are optimized by integrating data from ERP systems with qualitative information such as fashion trends and consumer trends surveys.
The case study performed by (Mostard, Teunter and De Koster 2011) is focused on improving the accuracy of demand forecasting procedures in the sector of fashion retailing and exploring the opportunities for mail order and Internet stores in the Netherlands. The study introduces advanced demand information-based forecasting methods to improve accuracy and efficiency. It compares these methods against existing methods, analyzing 700 SKUs over two summer seasons.
(Cannas et al. 2023) have presented a case study where AI is used in OSCM processes in six companies established in Italy. AI was used in 17 case studies with a target of measuring the applicability of AI in operations and supply chain management. SCOR and semi-structured interviews with the professionals of the industry who were involved in AI were applied through the Supply Chain Operations Reference (SCOR) model.