Sales Prediction using Ensemble Machine Learning Model

Abstract

With increased competition in the supermarket industry, there is an increased need for higher-order predictive analytics to
garner insight into consumer behavior for optimal sales strategies. Therefore, this research has presented a sales prediction
using an ensemble machine learning approach by considering multiple algorithms: Random Forest, XGBoost, and Support
Vector Machine, which further improve predictive accuracy and avoid possible overfitting. This paper presented a
comprehensive data preprocessing and feature engineering, with the implementation of a stacking ensemble model, which
resulted in excellent predictive performance. The stacking ensemble model achieved the best R2 value of 0.9990 and the least
mean absolute error. The results showed that machine learning techniques are very promising to improve sales prediction and
provide a powerful tool for supermarkets in making better decisions, optimizing inventories, and conducting focused
marketing. Hybrid models should be further explored in future research, with the addition of more external factors to improve
the predictive accuracy.