Enhancing House Price Forecasting with State-Action-Reward-State-Action Algorithm and Deep Q-Learning

The research aims to optimise housing pricing strategies using reinforcement learning algorithms, specifically SARSA and Deep Q-Network (DQN). The study employs the Kaggle House Prices dataset and focuses on variables such as year built, neighbourhood, and overall quality to enhance real estate price decisions. The SARSA algorithm was designed to dynamically change asking prices by providing incentives based on the difference between asking and actual selling prices, as well as a penalty for time on the market. This model fared well, with an average sale duration of 34.52 days and a total reward of around $1,640,667.27. The DQN algorithm used deep learning to find subtle patterns in the data in order to make better decisions. The DQN model outperformed SARSA, with an average total payout of $1,780,345.82 and a shorter average sale time of 31.75 days. The findings demonstrate how RL, specifically DQN, can be utilised to improve pricing strategies and provide valuable information about the real estate market. This study demonstrates the utility of sophisticated reinforcement learning approaches in complicated decision-making and identifies areas for further research into larger datasets and more complex models for better property price optimisation.