Housing has become one of the major challenges facing the world’s metropolises. In addition to the impact of macroeconomic changes, the characteristics of the housing stock itself constitute a clear driver of price changes (Bailey, Muth & Nourse, 1963). Variations in the quality of different homes and different characteristics make it difficult to estimate the price of a real property, so real estate appraisal reviews have played an increasingly important role in the real estate industry (Pavlov, 2000). There is a strong and continuing demand for techniques to determine the accuracy of appraisal reports from lenders, institutional investors, courts, and others who make decisions based on the veracity of appraisal reports (Benjamin, Guttery & Sirmans, 2004; Isakson, 1998). The value of multiple regression analysis and machine learning techniques in this area has been documented as a stand-alone technique to check the accuracy of appraisal reports (Isakson, 2001; Mak, Choy & Ho, 2010).
Therefore, we selected a proprietary collection of real estate data provided by the Chicago Cook County Assessor and other private and public organizations for analytical modeling to examine home price issues. The dataset includes homes sold in three areas of Cook County (North Township, City of Chicago, and Southwest Township) for tax years 2003-2018. We will examine the impact of different variables on property values in Chicago. This allows us to dig deeper into the variables and provide a model that can more accurately estimate home values. In this way, individuals and professional organizations can better price homes based on available information.
We use 20 explanatory variables that include almost all aspects of Cook County homes. Statistical regression models and machine learning regression models are applied and further compared based on their performance to better estimate the final price of each home. This projectBy applying various techniques from data analysis (such as linear regression and multiple linear regression) to its real residential property prices helps us understand the changes over years and potentially forecast future prices. Secondly, based on the house location we can do clustering (unsupervised learning) about Walkfac (Car-dependent, Walkable, Somewhat walkable), and we also check if there is a correlation between Walkscore and Moreover, we also want to know if the price per foot is related to the house located in Chicago. This project employs methods including correlation matrix, types of statistical tests, and OLS regression to analyze the random sample of homes that have sold in the three regions of Cook County (Northern Townships, City of Chicago, Southwest Townships) during the tax years 2003-2018 and to predict sale price to analyze the data and build an This project has big and detailed data with the data of the tax years 2003-2018.
Based on the data analysis results, we can see all the machine learning methods, the most important factor is location, then is building square feet, it quite makes sense. And then we can see the facility is second most important to the house price which including central air and fireplace, age, property class. At the same time, we have established a high-accuracy prediction model, and made high-accuracy predictions on housing prices based on 14 variables. Finally, based on unsupervised learning, we cluster the houses according to latitude and longitude and summarize the features of different categories.
This project has big and detailed data with high accuracy models, and we believe it provided us with very insightful results about the Chicago house market . These models estimate the implied price of each feature in the price distribution, so it can better explain real-world phenomena and provide a more comprehensive understanding of the relationship between housing characteristics and prices.