In today’s world, with the growing population and rapid economic changes, the issue of food security and optimal resource management has become more critical than ever. One of the fundamental challenges in this area is the management of food waste in the supply chain, particularly at the retail level. The increase in food waste not only leads to resource loss but also results in decreased profitability for retailers and causes social and environmental damage. Moreover, consumer behavior changes and the lack of accurate demand forecasting are among the factors that exacerbate food waste at the retail level. Therefore, it seems essential to develop dynamic solutions that can efficiently manage this waste. In this research, aimed at reducing food waste and improving the profitability of retailers, dynamic promotional and discount strategies were investigated. These strategies were optimized by considering parameters such as inventory, expiration date, and the economic condition of consumers, leading to models that predict both optimal discounts and food waste. For this purpose, real data were collected and analyzed from retailers in a specific geographic region. This data included information on inventory, sales, product expiration dates, and previously implemented discount policies. Preprocessing steps such as data cleaning, handling missing data, and normalization were carried out to prepare the collected data for the models. The "Buy One, Get One Free" (BOGOF) promotional strategy was then dynamically adjusted based on foundational research, so that the optimal policy could be recommended under various conditions, depending on factors such as inventory levels, shelf life, and retailer profitability. These strategies were designed to respond dynamically to changes in demand over time and prevent the generation of food waste. Next, various neural network models, including Long Short-Term Memory (LSTM), Bidirectional Long Short-Term Memory (BiLSTM), and Simple Recurrent Neural Networks (SimpleRNN), were implemented and eva‎luated to optimize discount and promotional policies. The goal was to reduce food waste and increase profitability across four product groups: bread and cereals, protein products, semi-prepared foods, and dairy products. These models were trained and optimized using real data collected from the studied stores to accurately predict food waste and determine the best time for applying discounts. The careful selection of hyperparameters for the neural network models, along with precise adjustments during training, significantly improved the accuracy of the models. Additionally, optimization techniques and variance reduction methods further enhanced the models' performance and minimized prediction errors. The results of the research showed that dynamic discount and promotional strategies not only improve retailers' profitability but also significantly reduce food waste. In the bread and cereals group, the model based on LSTM was able to predict optimal discounts and nearly eliminate food waste. Similarly, in the protein products group, the BiLSTM model minimized waste through the application of smart discounts. However, excessive discounts occasionally led to reduced revenue, which required more precise adjustments. In the semi-prepared foods group, the LSTM model successfully predicted demand and market conditions, resulting in a significant reduction in waste. Furthermore, the SimpleRNN model in the dairy products group showed satisfactory performance by reducing prediction errors and increasing profitability. Numerical analyses revealed that the proposed intelligent models outperformed traditional discount strategies, such as "Buy One, Get Two Free," by reducing waste and increasing profitability. These models, with their high adaptability and rapid response to demand fluctuations, allowed retailers to optimize their discount and promotional strategies effectively.

فرشته پرورش

مهدي علينقيان , صبا صارمي نيا