| dc.description.abstract | Housing prices are a crucial indicator of urban economic and social development. In recent years, housing prices in Taiwan have continued to rise, with Taipei City having the highest price-to-income ratio, posing challenges for real estate market participants in their decision-making processes. Although machine learning techniques have been applied to housing price prediction, previous domestic studies have had limitations such as incomplete training data coverage, lack of precise transaction coordinates, and failure to compare residential and commercial transaction data. This study aims to develop and evaluate machine learning models, comparing the features of price prediction between self-use and commercial residential buildings and apartments. We collected raw data on housing, geographic, and economic factors, and after data preprocessing, constructed a dataset with 45 predictive features, covering housing transaction data in Taipei City from 2016 to 2021, with a total of 44,918 records. The study employed five machine learning algorithms: Random Forest, Extreme Gradient Boosting, Adaptive Boosting, Neural Networks, and K-Nearest Neighbors. The models were trained using ten-fold cross-validation and evaluated with five performance metrics. The results showed that Extreme Gradient Boosting performed the best with an R² of 0.82. Further, regional partitioning reduced the prediction error, lowering the average MAE to 2.54 million NTD. The common key factors affecting price prediction were the region of the house and the construction date of the building. Important factors for predicting residential prices included the proximity of public facilities to the transaction target, while for commercial prices, the total transferred area was significant. | en_US |