1
CSC 573: Data Mining
Programming Project #2: Implementing Naïve Bayes Classification
Instructor: Ratko Orlandic
Your task is to implement two programs for Naïve Bayes classification. From the given data set
described below, the first program should produce a training set and a testing set after performing a
simple “randomization” of data (see below). The second program should derive the Naïve Bayes
classification model from the training set and evaluate its accuracy on the testing set (see below).
The programs can be developed in a programming language of your choice (Java, C, or C++), but
they should run on Windows. While this assignment text gives you some implementation freedom,
you should make every effort to follow this specification exactly.
Data Set
Your program must work on the tab-delimited data file “iris-discretized.txt” included in
Project2Files.zip. This is a version of the “iris” data set whose non-class attributes have been
discretized using equi-width binning. The file has 150 instances (data points), i.e. rows in the file.
The first 4 attributes (1. sepallength, 2. sepalwidth, 3. petallength, and 4. petalwidth) are assumed
to be nominal (values 1-5). Note that each value 1-5 represents a range of values in the original
“iris” data set from which I derived the “iris-discretized.txt” file. The last (5th) attribute is the class
dimension (nominal values 1-3 denoting types of iris). Note that a row in the data file represents a
data point (instance) and that a column represents an attribute (dimension).
Data Randomization
The first program should be called “randomize”. It should: a) read the given data file “iris-
discretized.txt”, b) perform a simple randomization of data points in it, and c) output two text files,
called “iris-train.txt” and “iris-test.txt”, which will later be used as the training and test data sets,
respectively. The program can be specific to the “iris-discretized.txt” data set and it should not take
any other input parameter.
For the purposes of “randomizing” the data, use the following simple procedure:
1) Read the “iris-discretized.txt” set into an array of data points.
2) Assume that the numbering of data points (rows) is between 0 and 149.
3) For each data point i (between 0 and 149) in the array:
3a) If i mod 3 = 0 or 1, include the data point i in the “iris-train.txt” set.
3b) If i mod 3 = 2, include the data point i in the “iris-test.txt” set.
Note that mod is the “modulo” operator that, in this case, computes the remainder of the integer
division of the number i by 3. This procedure will create a training data set with 100 data points
(instances) and a test set with 50 instances. You can create temporary “training” and “test” arrays
before writing the data points into the “iris-train.txt” and “iris-test.txt” files.
Data Classification
To implement the second program, called “classify”, please read relevant course slides (Part 4) and
section 6.4 in the textbook describing the Naïve Bayes classification. For the purposes of this
assignment, you should ignore the “zero-frequency problem” (i.e., assume that the computed
posterior probabilities can be 0.0). You also need not worry about the Naïve Bayes handling of
missing values and numeric attributes.
