Since the real estate investment trust (REIT) was first developed in the United States in 1960, more than 40 regions have established REIT markets. Similar to stock investments, investors in REIT also want to predict REIT trends. More recently, machine learning machine learning approaches have become some of the most popular methods for predicting stocks, helping investors to make a better decision. However, very few researchers have extended the implementation of machine learning approaches to REIT prediction and REIT-mixed portfolio management because the scale and capitalization of REIT market are relatively small compared to the stock market, especially in some developing countries, such as China. Nevertheless, REIT can provide stable cash flows to investors due to its mandatory requirement to distribute at least 75% annual income; REIT also delivers ongoing asset appreciation. REIT could thus be treated as an important financial product and asset in portfolio. The primary purpose of this thesis is to seek a better way for predicting REIT price and increasing REIT portfolios' return efficiently and accurately. [...]