Utilization of ABC/XYZ analysis in retail sales forecasting using multivariate LSTM method

Some significant developments that impact the retail sector are shifts in consumer demand and technology uptake. In order to provide greater convenience and improved customer experience, businesses are therefore compelled to streamline their business structures to match demand trends. That is only possible through leveraging efficient forecasting sales techniques that accurately predict future demand and enable proper reaction. Product classification based on ABC/XYZ analysis has become of high importance for minimizing technological costs when using complex algorithms. The study further proposes using a single Recursive LSTM model to estimate future sales using the categorized database, including daily previous sales records. Having a WMAPE of 17.61 % of the top five bestselling products, is still an outstanding result considering the global diversity of products and related variables. Besides knowing that the LSTM model can forecast long sequences, the innovative approach behind the presented technique is the key to the model robustness where certain inputs need to be carefully considered in order for the model to perform at its best ability. As the result, future sales projections closely resemble actual sales volume patterns. Thus, while incorporating a variety of economic variables like seasonality and calendar events, the model displayed a confident prediction accuracy.