Exploring Consumer Demand for Handloom Sarees in Kolkata

Demand forecasting plays a critical role in modern supply chain and inventory management, particularly in mitigating the risk of stock obsolescence and ensuring optimal resource utilization. Traditional statistical models often fail to capture the nonlinear patterns, seasonality, and volatility inherent in consumer demand. Deep learning, with its ability to learn complex temporal and contextual relationships, has emerged as a transformative approach to forecasting. This study explores the application of deep learning models—including Feed forward Neural Networks (FNN), Convolution Neural Networks (CNN), Recurrent Neural Networks (RNN), Long Short-Term Memory (LSTM) networks, Gated Recurrent Units (GRU), Transformer architectures, Auto encoders, and Hybrid models—for demand prediction in dynamic supply chains. Each model is evaluated based on its capacity to address critical challenges such as long-term dependency, data sparsely, anomaly detection, and adaptability to external factors. The research highlights the superiority of advanced architectures like LSTM, GRU, and Transformer-based models for capturing sequential demand fluctuations, while hybrid models integrating CNN with LSTM or Transformer layers demonstrate enhanced accuracy in multi-SKU and high-dimensional forecasting scenarios. The findings underscore that deep learning not only improves forecasting precision but also provides actionable insights for preventing excess stock, reducing obsolescence, and strengthening supply chain resilience.