TOBB ETU

9TOBB ETU
Aug 2023
Deadline:2021
6 Aug
BİL 470/570 Deadline
19 Aug 2023 23:59
HW 3 16 Aug 2021 23:59

Figure 1: k-means clustering on iris dataset

In this assignment you are going to implement a k-means clustering for classification.
You will train the model on the iris dataset (same dataset as in HW1) and interpret the
result of the classification while comparing the results with the decision tree classifier which
is the model in HW1.

1 Tasks
1.1 K-means Clustering Classifier
Train the K-means Cluster Classifier you have learnt in this course. To determine k value,
use the elbow method and plot ‘the number of cluster’ versus ‘sum of squared distances of
samples to their closes cluster center’. It should be like Fig. 2. To calculate distance between
samples and centroids, use Euclidean distance.
The signature of the aforementioned classifier will be as follows:

• KMeansClusterClassifier(n_cluster)

– fit(X, y)
– predict(X)

Train the classifier using the first %80 of the data and test it with the remaining data.
You cannot use any libraries to implement the KMeansClusterClassifier. It should
work with builtin types. For vectors you can use lists, and for 2D input you can use list of
lists.

1
 TOBB ETU
9TOBB ETU
Aug 2023
Deadline:2021
6 Aug
BİL 470/570 Deadline
19 Aug 2023 23:59
HW 3 16 Aug 2021 23:59

Figure 2: Elbow method for selection of optimal k cluster

1.2 Results
• Plot the 3D cluster plot as shown in Fig. 1

• Display confusion matrix

• Calculate following metrics:

– F1-Score
– Accuracy
– Precision
– Recall

• Plot the receiver operating characteristic (ROC) curve and calculate area under the
ROC curve (AUC)

• Comment on these results 1 .

• Compare these results with the output of decision tree classifier which is imple-
mented in HW1. Comparison should give the idea of that why one of them is better
than the other one, what is the advantages and disadvantages of using these methods,
in which situation which one is useful.

You can add additional plots or tables to elaborate results more.

You can use libraries like pandas, numpy, seaborn, matplotlib to show your results in this
part. Feel free to add additional libraries if you want.
1
All results should be given for both training and test data

2
 TOBB ETU
9TOBB ETU
Aug 2023
Deadline:2021
6 Aug
BİL 470/570 Deadline
19 Aug 2023 23:59
HW 3 16 Aug 2021 23:59

2 Submission
You are to submit 3 files:

1. Python file (kmeans.py): Contains the implementation of KMeansClusterClassifier.

It should have sufficent number of comments to explain the code you wrote. Using doc-
string is also plus. You cannot use any libraries in this file.

2. Notebook file (report.ipynb): Contains 2 part; (1) training of the classifier and,
(2) interpretation and comparison of the results. You can use markdown syntax to
explain steps, write python code to train the model, plot the graphs and tables.

3. Report (report.pdf): PDF export of the corresponding report.ipynb file. This file
should have same content with the notebook file. You can create this file from File >
Download as > .pdf from the menu of the jupyter notebook.

3 Notes for development environment

As you know, we will use python3 for this class. You need to setup few things in order to
get going:

• Pick a IDE (VSCode is suggested).

• Install python or install anaconda instead because you can use conda environments in
your project. Conda also contains python

• Install jupyter notebooks, if you install python via anaconda, this step can be ignored.

Feel free to ask questions on piazza.

Academic Integrity
This assignment is an individual assignment and cannot be done in groups. The originality
of your work should not be taken from a person or source. Demo for your assignments may
be asked nd your homework grade will be given based on your demo performance. If you
need supervision, you can apply to the assistant or the lecturer of the course. The homework
grade of students who are found to be cheating is considered 0 and a disciplinary measures
will be taken. In order not to put yourself and your friends in a difficult situation, you should
take the necessary care in homework.