Department of Computer Science and Software Engineering
MODULE CODE MODULE LEADER DEPARTMENT Phone
CSE315
2019/20 SEMESTER 1
BACHELOR DEGREE – YEAR 4 & MASTER DEGREE
MACHINE LEARNING
ASSIGNMENT 2
START DATE: 4 NOVEMBER
DUE DATE: 9 DECEMBER by 23:59
GENERAL INSTRUCTIONS
1. The assignment 2 consists of 3 tasks, which together are worth 100 marks. The
figure in [] denotes the number of marks available for that task or part of the
task.
2. The programs can be written in Matlab or Python in a professional style (e.g.,
appropriate comments, indents, meaningful variable names).
3. Please make sure that your programs are executable and your PDF report is
produced by computer.
4. No plagiarism, which will be strictly checked by Turnitin.
5. You are supposed to finish all the tasks.
6. Partial marks may be awarded depending on the degree of completeness and
clarity of your answers.
Department of Computer Science and Software Engineering
Task 1 Classification [35 Marks]
1. Using the MNIST database available at http://yann.lecun.com/exdb/mnist/, select two
classification algorithms and implement them to achieve a high accuracy (more than 90%).
[15 marks]
2. Describe the techniques, including data preparation, feature reduction, and training tricks in
your classification algorithms. [10 marks]
3. Analyse some other techniques that can be applied in your classification algorithms to
improve your model’s performance such as accuracy, efficiency, and storage. [10 marks]
Task 2 Support vector machine (SVM) and principal component analysis (PCA) [40
Marks]
1. Using the iris.data available on the ICE, select the training dataset and validation dataset, and
implement the SVM algorithm (based on public packages or libraries) to classify the types of
iris (achieving an accuracy of 90%). [10 marks]
2. Using the same iris.data, reduce the dimension of features applying the PCA and extract the
first, second, and third principal components. [10 marks]
3. Using the extracted first, second, and third principal component in Task 2.2, respectively, to
train a SVM model to classify the types of iris and compare their accuracies. [10 marks]
4. For each combination of the extracted first, second, and third principal components, train a
SVM model to classify the types of iris, and then compare their accuracies. [10 marks]
Task 3 Clustering [25 Marks]
1. Specify the number of clustering (e.g., 1, 2, 3, 4, 5, and 6) and implement the k-means
algorithm (based on public packages or libraries) to classify the iris.data. [10 marks]
2. Apply the PCA to reduce the dimension of features and combine the first, second, and third
principal components to implement the k-means algorithm (based on public packages or
libraries) to classify the iris.data. [15 marks]
END OF ASSIGNMENT 2