Forecasting superstore fruit and vegetable sales based on ARIMA-LSTM

Abstract

In the context of a fresh food supermarket, managing perishable vegetables, which have a notably limited shelf life, maintaining optimal stock levels becomes imperative for safeguarding profitability. This research paper intricately explores the analysis of numerous datasets using Power Query and Power BI, with a pivotal focus on devising a succinct coding system. To ensure data quality, outliers are meticulously eliminated using advanced techniques like linear interpolation and Z-Score processing. The ARIMA time series model is employed to predict the sales volume for each category in July, having been trained on data from June 2022 to June 2023, adeptly capturing historical sales patterns and emerging trends. Additionally, a parameter-tuned LSTM recurrent neural network, optimized with Adam's algorithm, is utilized to refine predictions, considering its inherent memory capabilities. A thoughtfully designed weighted average approach combines outputs from both models, strategically minimizing overfitting and enhancing prediction accuracy. The resultant research fortifies inventory management optimization, bolstering the supermarket's profitability while effectively meeting customer demands.