BIG MART PRICE ANALYSIS AND PREDICTION

Abstract

Nowadays, every company seeks to increase sales and maximize income, driven by meeting customer demand. Using historical data, machine learning models can help predict future outcomes. This research paper examines how machine learning can be utilized to enhance store sales. The models were applied to predict future prices, supporting inventory, and management decisions such as buying and selling stock. In addition, these models contribute to better business policies and help address real-world business challenges. The Big Mart dataset, collected in 2013, contains information on 1,559 products from 10 stores across different cities. Three regression models were applied, with the Gradient Boosting Regressor achieving the highest R-square value of 74%, demonstrating strong performance in price prediction. The KNN Classifier was also tested: on the non-null dataset, it achieved 85% accuracy, while on the dataset with missing values filled using the mode, it achieved 99% accuracy. However, this study has some limitations. The dataset does not include complete customer information, such as reviews, seasonal effects, or other behavioral factors. As a result, the findings from Big Mart stores may not be fully generalizable to other retail environments. Moreover, incomplete or poor-quality datasets can negatively affect prediction accuracy. For future research, collecting more customer-related information—such as shopping behavior, weather conditions (sunny, fair, rainy), reviews, visits during promotions versus non-promotions, and whether customers shop alone or with family—could improve both the accuracy and reliability of predictions.