Coursework: K-Nearest Neighbors

In the classification problem of two classes, two sets of vector characteristics are provided, one for each class.

The sets are stored in text files [url removed, login to view] and c2.dat.

Each file consists of lines and each line has 20 values (characteristics)

So a vector of characteristics is stored in each row.

A) Implement the classification algorithm k-NN, (k-Nearest Neighbors). The algorithm steps are:

1. Calculate the Euclidean distances of the classification vector, x, with all the vectors in

the first file ([url removed, login to view]).

2. Calculate the Euclidean distances for the classification vector, x, with all the vectors in

the second file ([url removed, login to view]).

3. Sort these Euclidean distances in ascending order (using both classes/files).

4. Assume an odd integer k, ie, k = 3. After the classification, count the k smallest Euclidean distances.

How many come from the first and how many from the second class?

Comment: A vector is classified in the Class with the most impressions.

B) Out of the 20 characteristics of the problem, select 3 (your choice).

From each data set select randomly 60% of the vectors as training vectors and use the rest to find

the classification error, according to the algorithm k-NN (k = 1 and k = 3).

Repeat the above process of random choice of the 60% of the vectors 10 times and calculate the average classification error for k = 3.

C) Instead of classifier k-NN, construct and train a neural network of your choice with the same 3 characteristics

you selected in (B). Then proceed as in (B) and comment/compare the

performance of the classifiers.

Inżynieria Java Linux Matlab i Mathematica Microsoft MySQL PHP Architektura oprogramowania Testowanie oprogramowania Pulpit systemu Windows

Numer ID Projektu: #3011024

O projekcie