Dynamic Pricing Strategy In Retail Using Deep Q-Learning And Genetic Algorithms

Abstract

Background: With e-commerce, dynamic pricing is a key strategy for maximizing revenue in retail. Conventional optimization and rule-based techniques are not real-time because the nature of consumer behavior and market conditions is volatile. Objective: The paper presents a hybrid intelligent pricing system based on Deep Q-Learning (DQL) and Genetic Algorithms (GA) to realize adaptive, autonomous, and cost-effective dynamic pricing in a retail environment. Methodology: The proposed model utilizes a Deep Q-Network (DQN) as the main decision-making module that learns the best pricing policies by engaging with a simulated retail market environment. The hyperparameters of the DQN, such as learning rate, discount factor, and the configuration of the network structure, are optimized by using the GA, which helps the algorithm to converge faster and prevent it from getting stuck in local optima. The state space is classified based on price elasticity indices, inventory levels, competitors' price signals, and temporal patterns of demand. The reward function is written to maximize profits and user conversion rate. Results: Electronic product transaction datasets were used for experiments, and the results show that the proposed hybrid DQL-GA model can improve the mean profit by 18.4%, the mean conversion rate by 12.7%, and the mean inventory turnover by 23.0% over the baseline rule-based method. The model also shows a performance better than that of standalone DQL and traditional optimization strategies on the basis of five performance metrics: Precision (98.9%), Accuracy (97.5%), Recall (96.5%), Area Under the Curve (98.0%), and Delay Reduction (4.9%). Conclusion: The proposed DQL-GA hybrid framework is scalable, robust, and interpretable for intelligent retail pricing and is shown to be resilient for stable trading, promotional peak, and overstock clearance scenarios.