Maximizing Supermarket Profits: Data-Driven Strategies for Pricing, Sales, and Forecasting

Abstract

The actual circumstances of daily life are crucial for the purchasing and pricing strategies of supermarkets. Developing strategies based on these circumstances can assist businesses in ensuring profits and fostering win-win cooperation. This paper explores methods to maximize profit through purchasing and sales strategies. Initially, the relevant data for various categories of vegetables is integrated. Through histograms, their sales patterns are directly understood, highlighting the most popular vegetables. Upon analyzing each vegetable category, it becomes evident that their sales data do not conform to normal distributions. Therefore, Spearman correlation coefficients are calculated, revealing strong correlations between certain categories, such as aquatic roots and edible fungi. A line chart depicting the top ten selling vegetables indicates a noticeable periodicity. Traditional fitting methods struggle to adequately model the sales of each vegetable category and their relationship with cost-plus pricing. To address this, additional factors such as holidays, weeks, and months are incorporated using techniques like random forest regression. This approach yields cost-plus pricing dependence curves that better capture the relationship, while effectively managing noise. Regarding sales volume prediction, the original data displays significant volatility, necessitating the handling of outliers using the threshold method. For missing data, linear interpolation is employed to mitigate the impact of continuous missing values on prediction accuracy. Subsequently, Adam-optimized long short-term memory (LSTM) networks are utilized to forecast incoming quantities for the next seven days. By extrapolating from normal sales volume, market capacity is estimated, allowing for additional sales through discount strategies. This framework has the potential to increase original income by 1.1 times.