台北市的房價議題時常成為大眾關注的焦點。一般來說,房屋成交價格取決於人們對過去交易價格的理解。在過去數十年中,許多研究人員透過電腦進行預測房價的研究。近年,一些研究也利用實價登錄資料集與不同的機器學習模型來分析、預測房價。 本研究利用實價登錄資料集,以線性回歸,多層感知器及LSTM長短期記憶模型進行房價預測,以不同的參數組合進行訓練並分析。實驗結果顯示,LSTM長短期記憶模型具有較好的預測結果。在優化器的選擇上,Adam優化函數表現出比SGD或RMSProp擁有更佳的效能。在本有限的實驗中,單層的深度神經網路比多層不同的深度神經網路具有較好的預測結果。
House price in Taipei city is a widely discussed issue. Generally, the price is decided by people’s understanding from past dealing-price. In the past decades, many researchers pay attention on the study of house price prediction by computer computation. Recently, different machine learning models are adopted to analyze the actual price registration dataset for predicting house price. This study examines linear regression, MLP (Multi-layer perceptron), and LSTM (Long Short-Term Memory) models on prediction of the actual price registration dataset. Various parameters and combinations are also test in our experiments. Experimental results show that LSTM deep neural network has better prediction than others. In the selection of optimizer, the Adam function exhibits better than SGD or RMSProp functions. In our limited experiments, single-layer deep neural network model leads to better results than different multi-layer deep neural network models.