1. Eco 6380 Predictive Analytics For Economists Spring 2016
Professor Tom Fomby
Department of Economics
SMU

2. Presentation 6 The K-Nearest-Neighbors Model For Prediction or Classification Chapter 7 in SPB

3. OUTLINE I. K-NN: A Nonparametric Method
A. No Parameters to Estimate as in Multiple Linear Regression
B. Definition of Euclidean Distance between Vectors
C. Recommendation: Standardize Input Variables before Proceeding

4. OUTLINE
II. Un-Weighted Nearest Neighbor Scores: Simple Average of Training Neighbor’s Output Values
III. Weighted Nearest Neighbor Scores: Weighted Average of Training Neighbor’s Output Values
IV. Therefore K-NN is a sophisticated “Step-Function” Predictor that relies on an Average of Neighborhood Output Values taken from the Training Data Set

5. OUTLINE
V. In the K-NN Prediction Problem the Neighborhood Size, K, is the “tuning” parameter
VI. In the K-NN Classification Problem there are two “tuning” parameters: The Neighborhood Size and the cut-off probability for choice selection
VII. The K-NN Tuning Parameters are Often Chosen so as to maximize the accuracy of scoring the Validation Data Set

6. Now for a Discussion of the Various Parts of this Outline go to the pdf file K-NN Method.pdf

7. Classroom Exercise: Exercise 4