Sales Prediction of Big Mart based on Linear Regression, Random Forest, and Gradient Boosting

Abstract

With rapid development of machine learning and data science approaches, many retailers are employing sales prediction to aid in strategy formulation and profit growth. However, it is challenging for some small merchants to access vast volumes of data for analysis. This paper investigates the viability of predict sales in a small-scale retail supermarket, and evaluates the performance of linear regression, random forest, and gradient boosting method in the corresponding limited dataset. The selected metrics for analysis are RMSE, MAE, and R2 score. Experiment results indicate that the linear regression model has a relatively large error and suffers from underfitting. The two decision-tree based models, random forest, and gradient boosting, perform similarly, with gradient boosting model outperforms by a small margin. This study illustrates that it is feasible to perform sales forecasting by machine learning techniques on a small collection of data. It also identifies the issues in this process and suggests potential fixed, giving small retailers a guideline for making effective sales predictions.