Real-Time Dynamic Pricing Strategies via Deep Reinforcement Learning Driven by Hybrid CNN–LSTM Demand Forecasting Engines

• Asher Noah

  covenant University

  Author

The convergence of deep learning-based demand forecasting and reinforcement learning (RL) has opened new frontiers for dynamic pricing in e-commerce, yet most existing frameworks treat forecasting and pricing as decoupled processes, resulting in suboptimal revenue outcomes. This study develops and validates a hybrid framework integrating a CNN–LSTM demand forecasting engine with a Deep Q-Network (DQN) pricing agent for real-time e-commerce applications. Using a curated dataset of historical transaction records, the forecasting module achieved 94.7% directional accuracy with a MAPE of 8.3%, while the DQN agent outperformed rule-based pricing baselines by approximately EUR 37,000 per annual episode (p = 0.004). Key predictors driving pricing decisions included inventory levels, demand fluctuations, and prior pricing behavior. The findings demonstrate that coupling accurate demand forecasts with reinforcement learning yields measurable revenue gains and establish a replicable framework for deploying integrated forecasting-pricing architectures. Practical implications include guidance on implementation barriers and the importance of explainability mechanisms for regulatory compliance.

• pdf

Copyright (c) 2026 Asher Noah (Author)

This work is licensed under a Creative Commons Attribution 4.0 International License.