ML Sales Prediction
A dynamic pricing system for an online retailer using predictions served by a multi-layered Feedforward Neural Network, a LightGBM regressor, and an Elastic Net, hosted on a containerized serverless architecture.
Problem Statement
Online retailers lose revenue through static pricing that fails to account for demand fluctuations, competitor pricing, and seasonal trends. A dynamic pricing system requires accurate sales predictions from multiple model perspectives to minimize risk and maximize margins.
Methodology
Built and evaluated three distinct models—a multi-layered Feedforward Neural Network, a LightGBM regressor, and an Elastic Net—to capture different aspects of sales patterns. Implemented a Flask-based API serving ensemble predictions. Deployed on a containerized serverless architecture for cost-efficient scaling.
Results
The ensemble approach outperformed any single model by capturing both linear and non-linear sales dynamics. Containerized deployment reduced infrastructure costs while maintaining sub-second prediction latency under load.