The Investigation of Rental Market Rent Models Based on Machine Learning

Abstract

In terms of the renting field, landlords often set rental prices subjectively based on market trends, lacking standardized criteria. This can lead to arbitrary pricing and potential market instability. In contrast, machine learning, with its advancements across various industries, offers a solution to model these complex, non-linear relationships. Although numerous studies have investigated the factors that impact rental prices, there is a scarcity of research dedicated to predicting rental prices, in contrast to the extensive body of work centered on forecasting property prices within the real estate sector. Firstly, this study examines key factors influencing rental prices. Utilizing a Kaggle dataset containing statistical information on Italian housing rentals, including architectural and geographical features, this study processes the data, handles missing values, removes outliers, and analyzes feature correlations. Various machine learning models, including Backpropagation (BP) neural networks, decision trees, random forests, gradient boosting regression, and multiple linear regression, are employed for rental price prediction. The study concludes by comparing model performance and discussing their strengths and weaknesses. Key findings reveal that neural networks outperform other methods in rental price prediction, showing the lowest Mean Absolute Error (MAE) and Root Mean Squared Error (RMSE) errors, while decision trees perform the poorest. Furthermore, bathroom count exhibits the highest correlation with rental prices, followed by property size, with villa status ranking third.