A Machine Learning Driven Dynamic Pricing Framework for Revenue Maximization in Online Retail Markets

Abstract

Retail pricing has grown into a complex challenge as market dynamics become increasingly volatile and consumer behaviour more unpredictable. Static pricing strategies, which set a fixed price regardless of demand fluctuations or seasonal trends, often fail to capture available revenue opportunities. This study proposes a machine learning driven dynamic pricing framework designed to maximize revenue in retail environments using the UCI Online Retail Dataset, which contains transactional records from 2010 to 2011. Five models were developed and evaluated: Linear Regression, Ridge Regression, Random Forest, Gradient Boosting, and a one-dimensional Convolutional Neural Network (1D-CNN). Demand forecasting was performed using rich temporal and statistical; lag features, rolling mean, rolling standard deviation, and calendar-based attributes. The dynamic pricing engine simulated candidate prices within historical bounds and selected the price that maximized the revenue function for each product. Results show that tree-based and deep learning models significantly outperformed linear approaches in demand prediction accuracy. The Gradient Boosting model achieved the lowest Root Mean Squared Error (RMSE) among classical models, while the CNN demonstrated strong generalization on test data. Dynamic pricing strategies produced measurable revenue improvements over static average pricing across all model types, with improvements validated through paired t-tests. These findings demonstrate the practical utility of combining demand forecasting with price optimization for retail revenue management.