Optimizations of Training Dataset on House Price Estimation

House price in many cases can strongly reflect local economy, predicting the real estate price can help consumer and the developers. In this way, consumers can inquiry price by giving a series of attributes which describes their ideal house and sellers can get the recommended price by uploading the property description. Thus, a fast, 24-hour responsive auto-inquiry system can greatly improve customers' and sellers' experience. To establish similar system, the dataset which is used to describe the sale of individual residential property in Ames is set as an experiment material. This paper discusses the influence from the data-preprocessing stage on the establishment of later machine learning model using different data groups. Later in the accuracy comparison, controlled experiment is implemented to show the accuracy among different organization of the algorithms. This paper is aimed at how much influence from the model choosing and data preprocessing stage will be made on the prediction accuracy and interpret methods used in preprocessing stage. Furthermore, the experiments show that the single model prediction's accuracy has on average 5% less than the stacked model. In addition, the results represented in this paper have the potential to be integrated into the housing sales system which can be applied to price delivery systems to benefit both consumers and sellers.