Scholarlink

This paper presents a production-leaning dynamic pricing engine for e-commerce that recommends profit-maximizing prices while respecting operational guardrails. The system learns a supervised demand model from real transaction data (Olist Brazilian e-commerce) and augments it with competitor prices (scraped with HTML caching) and exogenous demand signals (Google Trends with exponential backoff and local Parquet cache). Feature engineering integrates temporal effects, discounting behaviour, competitor pressure, inventory proxies, and trend signals. Price recommendations are computed via a constrained grid search maximizing (p − c) \(\hat{\text{D}}\) (p, x) with a volatility clamp (±5%). We compare tree-based regressors (LightGBM, XGBoost) on predictive accuracy (RMSE) and business outcomes (expected profit lift, elasticity diagnostics). The pipeline refreshes on a schedule and exposes a Streamlit dashboard; an evaluation module exports camera-ready tables and figures. We report results from a full run of the system and discuss implications for deployment and research.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Dynamic Price Optimization for E-Commerce Using Demand Prediction and Competitor-Based Adjustment

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Sruthy Nath,
  • Riya Agrawal,
  • Hetvi Bhora,
  • Swati Vaishnav

摘要

This paper presents a production-leaning dynamic pricing engine for e-commerce that recommends profit-maximizing prices while respecting operational guardrails. The system learns a supervised demand model from real transaction data (Olist Brazilian e-commerce) and augments it with competitor prices (scraped with HTML caching) and exogenous demand signals (Google Trends with exponential backoff and local Parquet cache). Feature engineering integrates temporal effects, discounting behaviour, competitor pressure, inventory proxies, and trend signals. Price recommendations are computed via a constrained grid search maximizing (p − c) \(\hat{\text{D}}\) (p, x) with a volatility clamp (±5%). We compare tree-based regressors (LightGBM, XGBoost) on predictive accuracy (RMSE) and business outcomes (expected profit lift, elasticity diagnostics). The pipeline refreshes on a schedule and exposes a Streamlit dashboard; an evaluation module exports camera-ready tables and figures. We report results from a full run of the system and discuss implications for deployment and research.