Location-based Score Prediction for Condominiums in Bangkok

Abstract

Condominiums are one of the most popular residential properties due to the resurgence of urban living. They are typically found in more metropolitan areas, such as the national capital region, offering easy access to restaurants, shopping, and various activities. Condominiums can be both profitable investment properties and enjoyable homes. However, finding the right one can be challenging. Based on previous work, one crucial factor is the location, which requires experience, expertise, and time to consider. Thus, in this study, we aim to: 1) investigate the factors that affect the potential location of a condominium, and 2) apply machine learning algorithms to create a prediction model for condominium scores. In this research, we collected a novel dataset comprising more than 2,000 condominiums in Bangkok, Thailand, and examined the location-based factors that influence the price and purchasing decisions of individuals. We extracted several features for model training and studied the variables that could be used to measure the potential of a condominium in numerical form. We proposed a multi-layer perceptron neural network with 5-Fold cross-validation and grid-search techniques. The experimental results demonstrate that our network achieves a mean squared error (MSE) of 0.0067 when using the density-based score labeling.