Valuations of residential properties using a neural network'

1
Valuations of residential properties using a
neural network.
Paper

• by Gary Grudnitski

2
Used By

• Mortgage Holder
• Prospective Lenders
• Government Officials

3
Objective

• Multiple Regression
• Neural Network

vs.
4
Problem Statement

• to mitigate the negative effects on estimates of
  property values due to imprecision in the
  specification of the valuation equation

5
Re-statement

• To improve the model.

6

7
Process

• Surf Web
• 8 variables
• Normalize Variables (0-1)
• Run Neural Net

8
Web Site
9
9 Variables

• House Size
• Lot Size
• Stories
• Pool / Spa
• Garage

• Selling Price
• Age
• Bedrooms
• Bathrooms

10
Neural Net
11
Training

• Train 119 Houses
• Training-Test 30 Houses
• Test Accuracy 100 Houses

12
Train-Test 30 Houses
13
Results
Average Error

• Multiple Regression 11.6
• Neural 9.5

14
Take Aways

• Good Project Outline

15
Credits

• Clothes by Nautica
• Hair by Patti

16
Questions/Facts
17
Popular Valuation Systems

• Regression Models

18
Normalization Equation

• I-norm (I min) / range

19
Logistic (sigmoid)

•                 1
• ai      -----------
•          1 e-neti/T

• e is the irrational number which starts 2.718
  ..., the base of natural logarithms

20
NN Software

• Shareware
• Written in C
• Roy W. Dobbins
• Eberhart and Dobbins 1990

21
Parser

• Written in C

22
Valuation Types

• 1. Appraisal Emulation/Expert System Theory of
  substitution as the guidelines to make
  adjustments to similar sales.
• 2. Multi-variant Statistical Correlation
  Statistical relationship among independent
  variables
• 3. Conformal Mapping and Price-time Indexing In
  this valuation approach, indexes of home price
  appreciation are built for the subject property
  and its immediate environment based on analysis
  of recent sales over the past three years.

23
Valuation Types

• 4. Solimars Neural Controller This is a
  complex, proprietary set of algorithms that
  weights the results of the valuation engines
  for the determination of the subject property
  value and the resultant value range. With each
  valuation, the neural controller determines which
  valuation methodology to place greater reliance
  upon, based on the quality and quantity of
  characteristic data that went into the analysis.