| Summary: | 碩士 === 國立臺灣科技大學 === 營建工程系 === 106 === Taiwan has become an aging society. To fulfill older adults’ economic and
everyday needs, a system of reverse mortgage has been developed and has become an
essential element of social welfare policies in Taiwan. This mortgage system is funded
by the government and managed by banks. The mortgage plan entails valuing the real
estate of potential beneficiaries by real estate appraisers; according to the valuation
results, decisions are made regarding the monthly payments. Therefore, constructing a
property valuation model to help real estate appraisers in executing property appraisals
and to provide a convenient reference tool for people undergoing this process is crucial.
Research has suggested that factors including favorable neighborhood amenities and
high accessibility to transportation facilities positively affect house prices. Studies have
constructed property valuation models using the linear distance between the coordinates
of a building and those of surrounding facilities as the model factors; however, in reality,
the walking distance between a house and a nearby amenity is not necessarily
equivalent to the linear distance between such two objects. Therefore, this study
converted the distances between a house and its surrounding amenities—namely a
subway station, department store, large park, neighborhood park, sports center, and
prestigious junior high schools and elementary schools—into actual walking distances.
The program for calculating the actual walking distance between two coordinate points
was developed using the Python programming language along with the Google Maps
Distance Matrix API. Model factors were determined using four approaches: all
selection, forward selection, backward elimination, and stepwise regression methods.
Two multiple regression analysis models were constructed for evaluating linear and
actual walking distances. These models were then trained using cross-validation
techniques, with the original sample data divided into testing and training datasets. The
III
performance levels of the two models in terms of the four regression approaches were
evaluated using the synthesis index. The results revealed that in all four regression
approaches, the model used for evaluating actual walking distances outperformed that
used for evaluating linear distances.
|
|---|
