Aim

The  main  objective  of  Assignment  2  is  to  conduct  predictive  analytics,  by  building predictive models on a  dataset using Python in the Jupyter Notebook environment.

This assignment will test your ability to:

●              Read and describe the data using basic statistics,

●              Split the dataset into training and testing,

●              Conduct multi-class classification using Support Vector Machine (SVM)**,

●              Evaluate and compare predictive models,

●              Explore different datasets and select a  particular dataset that meets certain criteria

●              Deal with missing data,

●              Conduct clustering using k-means

**  Not  taught  in  this  unit,  you  are  to  explore  and  elaborate  these  in  your  report submission.  This will be a mild introduction to life-long learning to learn by yourself.

Data

We will explore the following datasets in Part A (plus a dataset of your choice in Part B):

1.  FIT1043-Essay-Features.csv

2.  FIT1043-Essay-Features-Submission.csv

Format: each file is a single comma separated (CSV) file

Description: These two datasets were derived from a set of essays and are used to describe the essay features in numeric information.

Columns:

This data is pre-processed data on a set of essays that were provided on Kaggle.  You DO NOT have to download or process/wrangle the data from the original source.

Hand-in Requirements

Please  hand  in  the  following  4  files  (including  a  PDF  file  containing  your  code,  answers  and explanations to questions, a Jupyter notebook file (.ipynb) containing your Python code to all the questions, CSV file for your prediction in task A4 and the video file respectively):

●    The PDF file should contain:

o    Answers     and    explanations     to     the    questions.     Make    sure     to    include screenshots/images of the graphs you generat. Also, copy/paste your Python code to justify your answers for all the questions.

o    You  can  use  Microsoft  Word  or  other word  processing  software  to  format  your submission. Alternatively, generate your PDF from your jupyter notebook formatted using markdown. Either way save the final copy to a PDF before submitting.

●    The .ipynb file should contain:

o    A copy of your work using python code to answer all the questions.

●    The video file should contain:

o    A recording of yourself, explaining your answers to Part B.

o    You can use Zoom to prepare your recording.

o    Note  each student  is required to explain their approach in Part B only. Please see

Note:  Zip,  rar  or  any  other  similar  file  compression  format  is  not  acceptable  and will  have  a penalty of 10%.

Assignment Tasks:

Note: You need to use Python to complete all tasks.

Part A: Classification

A1. Supervised Learning

1.  Explain supervised machine learning, the notion of labelled data, and train and test datasets.

2.  Read the ‘ FIT1043-Essay-Features.csv’ file and separate the features and the label (Hint: the label, in this case, is the ‘score’)

3.  Use the  sklearn.model_selection.train_test_split  function  to  split  your  data for training and testing.

A2. Classification (training)

1.  Explain the difference between binary and multi-class classification.

2.  In preparation for classification, your data should be normalised/scaled.

a.  Describe what you understand from this need to normalise data (this is in your Week 7 applied session).

b.  Choose  and  use  the  appropriate  normalisation  functions  available  in sklearn.preprocessing and scale the data appropriately.

3.  Use the Support Vector Machine algorithm to build the model.

a.  Describe SVM.  Again, this is not in your lecture content, you need to do some self-learning.

b.  In  SVM,  there  is  something  called  the  kernel.     Explain   what  you understand from it.

c.  Write  the  code  to  build  a  predictive  SVM  model  using  your  training dataset.

(Note: You  are  allowed  to  engineer  or  remove  features  as  you  deem appropriate)

4.  Repeat Task A2.3.c by using another classification algorithm such as Decision Tree or Random Forest algorithms instead of SVM.

A3. Classification (prediction)

1.  Using  the  testing  dataset  you  created  in  Task  A1.3  above,  conduct  the prediction for the ‘score’ (label) using the two models built by SVM and your other classification algorithm in A.2.4.

2.  Display the confusion matrices for both models (it should look like a 6x6 matrix). Unlike the lectures, where it is just a 2x2, you are now introduced to a multi-class classification problem setting.

3.  Compare the performance of SVM and your other classifier and provide your justification of which one performed better.

A4. Independent evaluation (Competition )