A Hybrid Time-Series Sales Forecasting Model using LSTM and Neural Prophet with Pertinent Features

Sales prediction always depends on static factors like sales teams, sales management, promotional offers, pricing strategies, and marketing efforts. Sales prediction also depends on dynamic factors like macroeconomic conditions, natural disasters, market competition, consumer trends and behavior, seasonality, and technological advancements. Several existing works considered static attributes for sales prediction, whereas sales prediction always depends on dynamic attributes too, in real-time. Most of the existing works on sales prediction have not considered dynamic attributes. In this work, two dynamic attributes, like the change in trade policies due to the US-China trade war and global uncertainty, that affect sales prediction in India are considered and incorporated into the proposed Long Short-Term Memory (LSTM) and Neural Prophet (NP) hybrid model. The present work concentrated on two datasets containing electronics sales data and automobile sales data since the US-China trade war affected sales of electronics, pharmaceuticals, automobiles, textiles, footwear, and metals. A hybrid time-series forecasting model involving LSTM and NP is implemented in the present work. The proposed work was compared with different latest time-series forecasting models, and a comparative analysis was done. The proposed LSTM-NP achieved an accuracy of 96% on both electronics sales data and automobile sales data.