AdvaSell: An AI-Powered Dynamic Pricing System Using Linear Regression for E-Commerce Platforms

Vollständiger Abstract

Worum geht es in dieser Arbeit?

The development of dynamic pricing that considers the conditions of the market at the time of purchase has turned out to be one of the major tasks for online stores. In this paper, we present the implementation of AdvaSell, a price prediction system, which is based on linear regression and recommends product prices according to its category, brand, stock, history of purchases, and seasonality. On a random test subset of approximately 30,000 records from an e-commerce dataset, the model scored 0.9945 R², suggesting a high fit and stability of the result. We describe the data gathering and preprocessing pipeline, the six features that make up the model, the training and cross-validation process, and the Flask web app that serves predictions. We perform a comparison of our AdvaSell model against ridge regression and gradient boosting models (XGBoost) in terms of R², RMSE, MAE, MAPE and training time in order to evaluate the trade-off between precision, speed and transparency. We use linear regression as the main model, as in business evaluation of a price recommendation, the team usually requires explanation of the price, rather than a mere output from the black box.