Predicting apartment housing price from secondary data using XGBoost algorithm: Bangladesh Context

Abstract

The real estate sector in Bangladesh is experiencing rapid growth, driven by increasing housing demand, expanding middle-class populations, and rising per-capita income. Accurate property valuation remains a critical challenge due to numerous influencing criteria. This study introduces an Automated Valuation Model (AVM) for predicting apartment prices using the XGBoost machine learning algorithm, addressing the current lack of valuation options in online real estate marketplaces. The research collected data from a prominent online real estate marketplace comprising9,136 apartment listings across Bangladesh, with 8,123 listings from Dhaka. The methodology involved data preprocessing, including one-hot encoding of categorical variables, resulting in 574 independent variables. Bayesian optimization was applied to tune hyper parameters. enhancing model performance. Key hyper parameters included max_depth, eta, gamma, and sampling ratios. The AVM achieved an R2 of 0.91 and a Mean Absolute Percentage Error (MAPE) of 8.57% on an 80:20 train-test split. With 5-fold cross-validation, the R2 was 0.89, and MAPE was 9.09%, indicating robustness and reliability. Comparative analysis highlighted that the proposed model out performed several existing approaches in the literature. This research emphasizes the importance of feature selection, revealing that data quality directly impacts model accuracy. Future research recommendations include leveraging natural language processing for extracting data from text descriptions, integrating multiple online marketplaces, and incorporating image processing techniques to enhance apartment price prediction. The proposed AVM offers a scalable solution for accurate, automated apartment price prediction, benefiting buyers, sellers, and financial institutions in Bangladesh's burgeoning real estate market using machine learning.