1 Bayesian, Decision Tree and Dependence Tree Classi ers
In this assignment, you will be implementing a few classification algorithms including the optimal Bayesian classier, one for Decision Trees (DTs), and one for Dependence Trees, and using them to classify several di event data sets.
Binary-valued Artificial Data Sets
Use the scheme below to generate the data sets you need:
You are dealing with a d-dimensional feature space with c = 4 You can assume that d = 10.
Assume that the vector components obey a Dependence Tree structure between the various features. This Dependence Tree must be arbitrarily assigned and unknown to the classification (i.e., training and testing)
For each of the c classes and for each of the d features, randomly generate the probabilities of the feature taking the value 0 or 1. Thus, for class j = 1; : : : ; c and for feature indices i = 1; : : : ; d, you must randomly assign the value vi;j = P r[xi = 0j! = !j]. These values must be based on the Dependence Tree that you have chosen.
Generate 2,000 samples for each class based on the above
Training and Testing
With regard to training and testing, do the following:
Use a 5-fold cross-validation scheme for training and
Using estimates of the vi;j's, estimate the true but unknown Dependence Record the results of how good your estimate of the true but unknown Dependence Tree is.
Perform a Bayesian classication1 assuming that all the random variables are independent. Notice that in this case, you must not assume a Gaussian distribution for the features, but the binary
Perform a Bayesian classification assuming that all the random variables are dependent based on the dependence tree that you have
Perform the classification based on a DT algorithm. For the DT algorithm, have your program output the resulting The output2 should be neatly indented for easy viewing.
1 Each data set has more than two classes. In each case, you must do the classification using a pairwise classification on all the classes and assign the testing sample to the most appropriate winning class. This paradigm must be followed for the other classification tasks too.
2 An excellent program to draw decision trees is Graphviz, available at: http://www.graphviz.org/.
Binary-valued Real-life Data Sets
In this section you will deal with the one Real-life data set.
The Glass Identi cation data set3 is to be used to classify the type of glass, given the following features, speci ed in this order:
Class: In this case there are 7 possible types, which can be further split in to 2 categories of windowed and non-windowed glass
RI: Refractive index
Na: Sodium (unit measurement is weight percent in the oxide, as are attributes 5-11)
You may ignore all the features that are non-numeric. Whenever you need binary features (i.e., for training and classifying using the Dependence Tree and Decision Tree), render the features to be binary by adopting a thresholding mechanism.
Techniques to be Implemented
Perform all the tasks given in Section 1.2.2 on this real-life data set.
Write a 2-3 page report summarizing all your results. The report should be relatively
Compare the classi cation accuracy of the Dependence Trees you have obtained for the arti cial and real-life data
Compare the classi cation accuracy of the four algorithms for the arti cial data sets. Do some seem to outperform others? Discuss the possible reasons for these
Compare the classi cation accuracy of the four algorithms ((a) Bayes, (b) Naive Bayes,
(c) using Dependence trees, and (d) using Decision Trees) for the real-life data sets. Do some seem to outperform others? Again, discuss the possible reasons for these results.
The ready solutions purchased from Library are already used solutions. Please do not submit them directly as it may lead to plagiarism. Once paid, the solution file download link will be sent to your provided email. Please either use them for learning purpose or re-write them in your own language. In case if you haven't get the email, do let us know via chat support.
Get Free Quote!
256 Experts Online