1# RECORD ba6d3043dcfb0e7c6ad45ae35c0629afbc728c5f2# TITLE: An ANN Model for the Influence of Siding Materials on Single-Family Home Values3# YEAR: 2003.0 VENUE: nan DOI: 10.18260/1-2--125494# AUTHORS: Mohammed E. Haque; A. Taibah5# PAGES: 9 CANDIDATE TABLE PAGES (1-based): [5]6# ABSTRACT: The real estate valuation is a complex process considering the range of variables that are known to play a role in determining such a value. The esthetics and choice of building’s facade materials can greatly influence a homebuyer’s decision. This paper evaluated such buyer’s preferences of various siding materials by estimating the market-clearing prices determined by the buyers of single-family homes (SFHs). The study employs a comparative hedonic estimation and artificial neural network (ANN) model. The study relied on a large sample of SFHs that were sold during the period from 1997 to 2000 in College Station, Texas. The analysis was restricted to homes that are highly homogeneous in their structural attributes in order to eliminate the impact of such attributes on the home values. The main aim of this study was to develop an ANN model to determine the influence of four different siding materials on home values. Several different ANN models with different layers/slabs connections, weights, and activation functions were trained. The presented ANN Jordan-Elman Nets with logistic activation function was the best one among all other trials. It converged very rapidly to produce a very high predictive efficacy. The trained model was evaluated using the data that was not used during the training, which also indicated very good agreement between the actual and ANN predicted price. Additionally the study was aimed to compare the hedonic value estimation and ANN value prediction methods. The results indicate that stucco siding had the most significant impact on the property value. The estimated implicit values of different siding materials were a measure of the importance of such materials to the homebuyer and was resembled in the form of a paid premium. Moreover, in comparing the hedonic results and ANN results, this study found that both analytical methods support one another and have assigned similar weights to the various construction types that have been studied. In addition ANN showed to have a higher predictive accuracy level than did the hedonic estimation. The findings extend the body of literature concerned with real estate value analysis and have significant implications in the realm of fund allocation decision making for a real estate developer.789===== PAGE 1 (score 6, numeric-paren 0, table-words 0, var-words 6) =====1011Session 1106 12Proceedings of the 2003 American Society for Engineering Education Annual Conference & Exposition 13Copyright 2003, American Society for Engineering Education 14 15An ANN Model for the Influence of Siding Materials 16on Single-Family Home Values 17 18 19Mohammed E. Haque, Ph.D., P.E., Amro Taibah, Ph.D. 20 21Texas A&M University, Texas, USA/ King Abdulaziz University, Saudi Arabia 22 23 24 25Abstract 26 27The real estate valuation is a complex process considering the range of variables that are known 28to play a role in determining such a value. The esthetics and choice of building’s façade 29materials can greatly influence a homebuyer’s decision. This paper evaluated such buyer’s 30preferences of various siding materials by estimating the market-clearing prices determined by 31the buyers of single-family homes (SFHs). The study employs a comparative hedonic estimation 32and artificial neural network (ANN) model. The study relied on a large sample of SFHs that were 33sold during the period from 1997 to 2000 in College Station, Texas. The analysis was restricted 34to homes that are highly homogeneous in their structural attributes in order to eliminate the 35impact of such attributes on the home values. The main aim of this study was to develop an ANN 36model to determine the influence of four different siding materials on home values. Several 37different ANN models with different layers/slabs connections, weights, and activation functions 38were trained. The presented ANN Jordan-Elman Nets with logistic activation function was the 39best one among all other trials. It converged very rapidly to produce a very high predictive 40efficacy. The trained model was evaluated using the data that was not used during the training, 41which also indicated very good agreement between the actual and ANN predicted price. 42Additionally the study was aimed to compare the hedonic value estimation and ANN value 43prediction methods. The results indicate that stucco siding had the most significant impact on the 44property value. The estimated implicit values of different siding materials were a measure of the 45importance of such materials to the homebuyer and was resembled in the form of a paid 46premium. Moreover, in comparing the hedonic results and ANN results, this study found that 47both analytical methods support one another and have assigned similar weights to the various 48construction types that have been studied. In addition ANN showed to have a higher predictive 49accuracy level than did the hedonic estimation. The findings extend the body of literature 50concerned with real estate value analysis and have significant implications in the realm of fund 51allocation decision making for a real estate developer. 52 53Introduction 54 55Artificial intelligence (AI) applications have gained a broad interest in civil/construction/ 56architectural engineering problems. Its applications are very extensive and interdisciplinary. The 57graduate students civil/construction/architectural students should especially be encouraged to 58learn various applications of computing techniques including artificial neural network (ANN), 59genetic algorithm (GA), etc. This paper highlights various applications of AI. As an example of 60Page 8.181.1616263===== PAGE 5 (score 30, numeric-paren 5, table-words 8, var-words 9) =====6465Proceedings of the 2003 American Society for Engineering Education Annual Conference & Exposition 66 Copyright 2003, American Society for Engineering Education 67Results of Hedonic Procedure 68 69An analysis of the data was performed using the hedonic estimation procedure in order to 70ascertain the relationship between property market value and the type of siding material used. 71The hedonic price function describes the relationship between the observed prices of 72commodities and the characteristics associated with the commodities, based on the hypothesis 73that “goods are valued for their utility-bearing attributes or characteristics,” 12. Hence the 74coefficient estimates represent the premium paid by the homeowner for the aesthetic utility that 75any siding material might bare. Based on the results of the analysis, the hedonic equation can be 76written as follows: 77 78PPSQFT00 = 52.419 - 0.515*AGE_YR + 16.595*CONST_VENEER + 15.734* CONST_FRAME + 7918.323*CONST_MIX +37.642*CONST_STUCO + e 80 81 82 83Where e = error term. 84 85The F-value of the model was found to be statistically significant at the 0.0001 levels. The F 86statistic basically tests how well the model, as a whole, accounts for the dependent variable's 87behavior. The estimation efficacy of the model was found to be moderately high with a 88coefficient of determination (R2) of 0.622. This is an indication that the model is able to explain 89approximately 62% of the total variation in the property’s market value. The results of this study 90indicate that “construction type” is an important attribute that has a significant effect on SFH 91market value. Moreover the findings indicate that certain types of construction have a more 92considerable impact than others. For example, based on the study sample, it was found that using 93various types of stucco siding in exterior finishing could have a noticeably higher impact on 94future market price than other sidings. Compared to brick veneer, which is the most common 95exterior finishing material in the College Station market, Stucco finishing was found to reflect an 96increase difference of approximately $21 per square foot in market price. In addition it was 97found that plywood sidings had the least premium estimated at approximately $15 per square 98foot. Homes with mixed siding materials in the sample were mainly brick veneer mixed with 99either wood or stucco sidings. This category also showed an estimated hedonic price of 100approximately $18 per square foot. 101 102ANN Model and Results 103 104The parameter inputs considered in the proposed ANN model were the ones that were found to 105be statistically significant using the hedonic procedure (AGE_YR, CONST_VENEER, 106CONST_FRAME, CONST_MIX, CONST_STUCO). The output parameter was price per square 107foot adjusted to 2000 value (PPSQFT00). During training the model, 130 observations were 108used out of a total of 145 observations. The remaining 15 observations, separated arbitrarily 109from the total number of observations, were used during testing the trained model. Several 110different ANN models with different layers/slabs connections, weights, and activation functions 111(including logistic, linear, Symmetric Logistic, and Gaussian) were trained. In addition, pattern 112selections including "Rotation" and "Random" were applied with weight updates using Vanilla, 113Momentum and TurboProp. The presented ANN- Jordan-Elman recurrent network with linear 114and logistic mixed activation function ("Rotation" for pattern selection, and "Momentum" for 115weight updates) was the best one among all other trials. In this research, Jordan-Elman recurrent 116network converged very rapidly to produce a very high predictive efficacy. Fig. 3 shows the 117Page 8.181.5118119120===== PAGE 6 (score 11, numeric-paren 0, table-words 3, var-words 5) =====121122Proceedings of the 2003 American Society for Engineering Education Annual Conference & Exposition 123 Copyright 2003, American Society for Engineering Education 124-40125-2012601272012840129601308013110013212013311346135111361613721138261393114036141411424614351144561456114666147711487614981150861519115296153101154106155111156116157121158126159Pattern Processed160Price $ per Sq. Ft.161Actual Price162ANN Predicted Price163Errors164Fig. 3 Actual and ANN Predicted Price during the Network Training Phase165016620167401686016980170100171120172117321743175417651776178717981809181101821118312184131851418615187Pattern Processed188Price $ per Sq. Ft.189Actual Price190ANN Predicted Price191Fig. 4 Actual and ANN Predicted Price during the Network Evaluation192actual and ANN predicted price ($ per Sq. Ft.) during the Network Training Phase, which 193indicates very low errors between the actual price and the ANN predicted price. The trained 194model was evaluated 195using the data that was 196not used during the 197training. Fig. 4 depicts 198the actual and ANN 199Predicted Price during 200the network evaluation 201phase, which indicates 202very small difference 203between the actual and 204ANN predicted price. 205During network training, 206R2 was obtained as 2070.7788, and 0.8627 208during network 209evaluation, which were 210very close to 1.0, 211indicating a very good fit 212between the actual and the 213network prediction. R2 is a statistical indicator usually applied to multiple regression analysis, 214and was calculated using the following formulae13: 215R2 = 1 - (SSE/SSyy) 216 217Where 218SSE = Σ (y - y)2, 219SSyy = Σ (y - y)2, 220y is the actual value, 221y is the predicted value of y, 222and y is the mean of the y 223values. 224 225 226ANN Price Prediction 227 228Hypothetical data was 229developed in order to 230compare the performance of 231the hedonic estimation 232model and the ANN model. 233The purpose is to further the 234validation and evaluate the 235performance similarities or 236differences between the two methods. Ten imaginary data points were used with building ages of 2375 and 10 years old and siding material characteristics that were chosen arbitrarily. The two 238Page 8.181.6239