E-Commerce Price Prediction and Analysis Using XGBoost

This paper presents OmniPrice, a full-stack e-commerce price intelligence system that combines real-time web extraction,
machine learning-based price forecasting, and interactive data visualisation within a unified serverless-capable platform. The system
integrates a FastAPI backend equipped with Playwright-driven headless browser scraping, a LangChain–Groq AI agent for
structured data extraction, and an XGBoost regression model trained on engineered temporal and categorical features for short
horizon price prediction. A Streamlit-based frontend dashboard renders live price snapshots, 30-day historical trends, and forecast
outputs via Plotly visualisations. Persistent price history is stored in SQLite by default with optional PostgreSQL support; Redis
provides a transparent caching layer with graceful fallback. Docker Compose orchestration packages all four services — database,
cache, backend, and frontend — into a reproducible deployment target. Evaluation across representative e-commerce product pages
from Amazon India, Flipkart, Myntra, and Croma demonstrates end-to-end extraction latency below six seconds at median,
XGBoost forecast MAE of 2.3% relative to reference prices, and Redis cache hit rates above 80% under repeated query loads.
Hardware and infrastructure cost is negligible compared to commercial price-tracking services, establishing economic feasibility
for individual researchers and small retail analytics teams.