Dynamic Pricing using Reinforcement Learning in Hospitality Industry

Abstract

Hotel room pricing is a very common use case in the hospitality industry. Such use cases take dynamic pricing strategies for setting optimum prices wherein prices are dynamically adjusted based on user engagement. However, it is challenging to design an approach that makes pricing dynamic with respect to complex market change. In this paper, we suggest a reinforcement learning based solution for this problem. The approach employs a Deep Q-Network (DQN) agent trained to recommend/suggest optimum pricing strategies which maximizes the total profits for a day. In addition, the pricing strategy is optimized in such a way that empty rooms remain minimal. A real-life hotel-bookings data set is being used for testing this approach. The data is aggregated and preprocessed before being used for the task. The pricing strategy is influenced by the hotel-demand, type of rooms, number of nights and other variables. The hotel-demand is derived from a Random-forest model trained on the processed data to simulate original demand distribution of processed data. Using the DQN based dynamic pricing strategy, a potential 15–20 percentage higher reward(profits) were obtained compared to fixed pricing, and rule-based pricing strategy. At the same time the empty rooms left were significantly lower for the DQN based approach.