دانلود مقاله ISI انگلیسی شماره 6781
ترجمه فارسی عنوان مقاله

ارزیابی توده دارایی واقعی توسط محاسبات هوشمندانه

عنوان انگلیسی
The mass appraisal of the real estate by computational intelligence
کد مقاله سال انتشار تعداد صفحات مقاله انگلیسی
6781 2011 6 صفحه PDF
منبع

Publisher : Elsevier - Science Direct (الزویر - ساینس دایرکت)

Journal : Applied Soft Computing, Volume 11, Issue 1, January 2011, Pages 443–448

ترجمه کلمات کلیدی
حداقل مربعات معمولی - رگرسیون بردار - چند لایه - کمیته
کلمات کلیدی انگلیسی
پیش نمایش مقاله
پیش نمایش مقاله  ارزیابی توده دارایی واقعی توسط  محاسبات هوشمندانه

چکیده انگلیسی

Mass appraisal is the systematic appraisal of groups of properties as of a given date using standardized procedures and statistical testing. Mass appraisal is commonly used to compute real estate tax. There are three traditional real estate valuation methods: the sales comparison approach, income approach, and the cost approach. Mass appraisal models are commonly based on the sales comparison approach. The ordinary least squares (OLS) linear regression is the classical method used to build models in this approach. The method is compared with computational intelligence approaches – support vector machine (SVM) regression, multilayer perceptron (MLP), and a committee of predictors in this paper. All the three predictors are used to build a weighted data-depended committee. A self-organizing map (SOM) generating clusters of value zones is used to obtain the data-dependent aggregation weights. The experimental investigations performed using data cordially provided by the Register center of Lithuania have shown very promising results. The performance of the computational intelligence-based techniques was considerably higher than that obtained using the official real estate models of the Register center. The performance of the committee using the weights based on zones obtained from the SOM was also higher than of that exploiting the real estate value zones provided by the Register center.

مقدمه انگلیسی

There are mass and individual appraisals of real estate. The individual appraisal is such an appraisal, when value of the exact object is determined according to all its individual characteristics. Mass appraisal is the systematic appraisal of groups of properties as of a given date using standardized procedures and statistical testing [40]. This valuation method is applied to property objects with many similarities. Mass appraisal of real estate is commonly applied to compute real estate tax. The purpose of mass valuation is to estimate the market value. It must be distinguished from the market price and other, non-market values [17]. According to the Lithuanian Republic normative documents, the market value is estimated as the money amount, for which a property can be exchanged on the valuation date between a willing buyer and a willing seller in arm's-length transaction after proper marketing, wherein the parties act knowledgeably, without compulsion and impact of other transactions and interests [27] and [28]. In the international valuation standards 2005 (IVS), issued by the International Valuation Standards Committee (IVSC), the market value is defined as the estimated amount of money for which a property should exchange on the date of valuation between a willing buyer and a willing seller in arm's-length transaction after proper marketing wherein the parties acted knowledgeably, prudently and without compulsion [17]. The market price is formed when curves of supply and demand intersect, it is influenced by many objective and subjective factors. The market price equals to the market value very rarely, because the market of real estate is not an ideal market. The market price of real estate reflects many subjective factors, so a real estate assessor must find the most objective, suitable for all value. There are three traditional real estate valuation methods: the sales comparison approach, income approach, and the cost approach [30] and [40]. According to the sales comparison approach, the value is determined by comparing the object with the other objects sold in the market. The value is adjusted according to differences, as real estate objects have the differences. A difference up to 30–35% between characteristics of various objects is acceptable [26]. This method is very suitable for clear land. Reflection of the market price, quick and simple computations is the main advantages of this approach. The income approach is based on the premise that the value is the present worth of future; the value is determined by discounting cache flows generated by the object. The approach is very suitable for objects generating incomes, for example, buildings with leased offices or flats, objects used for services or production. This approach is quite simple too and estimates the economic benefit from the object. In the cost approach case, the value of the object is determined by construction costs minus depreciation. This approach can be applied only to buildings, and it is very suitable for schools, objects of engineering infrastructure and similar, which do not generate incomes and there are only a few objects to compare with. Mass appraisal models are commonly based on the sales comparison approach. The linear regression is the most popular technique used to build mass appraisal valuation models. However, other techniques such as neural networks, support vector machines (SVM), and committees of models can also be used. Most studies on this topic compare the performance of linear regression and neural network-based models. One of the first well known studies was performed by Do and Grudnitski [37]. The authors used a one hidden layer perceptron trained by backpropagation. A very small 6.9% mean absolute error was achieved using sales data of individual houses in San Diego at 1991. The mean absolute error achieved with the linear regression on the same data was equal to 11.26%. Similar results were achieved in other studies by Borst [38], Tay and Ho [11], and Evans et al. [1]. However, Worzala et al. tested the previous studies with similar data and their results were not so promising [12]. For example, while trying to replicate the study of Do and Grudnitsky, for the neural networks they achieved only 10% and 13.1% mean absolute errors using different software packages, while the mean absolute error of linear regression was equal to 11.1%. The conclusion, therefore, was that neural networks must be used very carefully for the real estate valuation. There were many more studies by the other authors: for example, Amabile and Rosato [39], Rossini [32] and [33], Nguyen and Cripps [31], Ge et al. [21], Wilson et al. [16]. Their results show slight advantage of neural networks against the linear regression. It is obvious that both techniques may show advantage against each other depending on the quality and amount of the data, dependencies between the variables. The purpose of this study is to explore the usefulness of the most prominent computational intelligence techniques for mass appraisal. The linear regression which is the most popular technique in various studies, a multilayer perceptron (MLP), a support vector machine, and a committee of the models are the techniques used in the investigation. The number of unacceptable valuations is the main parameter of usefulness of a mass appraisal technique. Valuation differing from real value of an object more than certain percentage, 20% in Lithuania [28], is called an unacceptable valuation. There are two main reasons of using the linear regression instead of some higher order regression. Firstly, the linear regression is still used in most mass appraisal systems. Thus, old mass appraisal models can be included into a committee and, therefore, a much easier way of moving from legacy appraisal systems to ones proposed in this work is created. Secondly, Worzala has demonstrated that the linear regression is sometimes more accurate than non-linear computational intelligence techniques [12]. Our results also confirmed this fact. SVM and committees of models have shown excellent performance in the recently studied task of detecting fictious real estate transactions [46], [47] and [48]. We present a way of obtaining data-dependent weights for aggregating the models into a committee. The weights are based on real estate value zones obtained from the SOM. We demonstrate that committees based on value zones generated by SOM are more accurate than those exploiting the real estate value zones provided by the Register center. The finding means that the time consuming and expensive step of establishing the real estate value zones by an expert can be avoided. The remainder of the paper is organized as follows. In the next section, a brief description of the techniques used to solve the task is given. The results of the experimental tests are presented in Section 3. The conclusions of the study are given in Section 4.

نتیجه گیری انگلیسی

Ordinary (OLS regression) and computational intelligence-based techniques (MLP, SVM, and a committee comprised of all the three types of models) have been evaluated in the mass appraisal of real estate task. Many previous authors reported diverse results when comparing the MLP and OLS regression based approaches. Our results also indicated the superiority of OLS regression over the MLP. However, the SVM clearly outperformed both the OLS regression and MLP based models. The results indicate that non-linear modeling is required. SVM as being capable of non-linear modeling and finding the global minimum of the cost function suits very well for the task. The proposed committee of models has shown an excellent performance and clearly outperformed the separate predictors. The number of unacceptable valuations, which is the main parameter in the mass appraisal tasks, was only 1. It means that only 1% of valuations do not satisfy the accuracy limits for the mass appraisal. A new approach to obtaining data-dependent aggregation weights to build a weighted averaging committee of models has been presented in this work. Two ways of obtaining the weights were explored, namely the weights based on value zones obtained from experts of the Register center and the weights based on clusters (zones) provided by SOM without using any information from the experts. Clusters obtained from the Register center and SOM were quite different. This can be explained by the homogeneity of the whole area. Thus, the impact of house place was not so important in comparison to its main characteristics. The committee results obtained using SOM clusters were slightly better than those based on the Register center value zones. Therefore, SOM can be used for effective clustering of the real estate data. The much better performance obtained from the computational intelligence-based techniques than from the official models shows the great practical impact of this study.