A Combinatorial Deep Learning and Deep Prophet Memory Neural Network Method for Predicting Seasonal Product Consumption in Retail Supply Chains

Abstract

One of the biggest problems in supply chain networks is demand forecasting. It was created to increase demand, profitability, and sales while maximizing stock efficiency and cutting costs. To improve demand forecasting, historical data may be analyzed using a variety of techniques, such as deep learning models, time series analysis, and machine learning. This study develops a hybrid approach to demand prediction. This paper used a deep learning-based Deep Prophet memory neural network forecasting approach, which combined temporal, historical, trend, and seasonal data, to develop a more accurate model. To our knowledge, this is the first study to integrate prophet with a long short-term memory network (LSTM) for demand forecasting. At first, the data may be obtained here, and linear clipping data normalization (LCDN) was used for pre-processing. After that, the bivariate wrapper forward elimination approach may be used to extract the features. The Sequential Bayesian Inference Optimization (SBIO) approach is used to choose the specialized features from the retrieved characteristics. Ultimately, the demand for the items was examined after the Deep Prophet Memory Neural Network (DPMNN) approach was modified. Using M5 Forecasting and Predict Future Sales datasets in a Python context, the built system is used and evaluated. Numerous exhaustive and comparative trials show that the suggested demand forecasting technique performs much better than state-of-the-art research.