Creating your data set

Clear your workspace, set the number of significant digits to a sensible value, and use ‘WAUS’ as the default data frame name for the whole data set. Read your data into R and create your individual data using the following code:

rm(list = ls())

WAUS <- read.csv("HumidPredict2023D.csv")

L <- as.data.frame(c(1:49))

set.seed(XXXXXXXX) # Your Student ID is the random seed

L <- L[sample(nrow(L), 10, replace = FALSE),] # sample 10 locations

WAUS <- WAUS[(WAUS$Location %in% L),]

WAUS <- WAUS[sample(nrow(WAUS), 2000, replace = FALSE),] # sample 2000 rows

Questions (10 Marks)

1. Explore the data: What is the proportion of days when it is more humid than the previous day compared to those where it is less humid? Obtain descriptions of the predictor (independent) variables – mean, standard deviations, etc. for real-valued attributes. Is there anything noteworthy in the data? Are there any attributes you need to consider omitting from your analysis?

2. Document any pre-processing required to make the data set suitable for the model fitting that follows.

3. Divide your data into a 70% training and 30% test set by adapting the following code (written for the iris data). Use your student ID as the random seed.

set.seed(XXXXXXXX) #Student ID as random seed

train.row = sample(1:nrow(iris), 0.7*nrow(iris))

iris.train = iris[train.row,]

iris.test = iris[-train.row,]

4. Implement a classification model using each of the following techniques. For this question you may use each of the R functions at their default settings if suitable.

• Decision Tree

• Naïve Bayes

• Bagging

• Boosting

• Random Forest

5. Using the test data, classify each of the test cases as ‘more humid tomorrow’ or ‘ less humid tomorrow’ . Create a confusion matrix and report the accuracy of each model.

6.Using the test data, calculate the confidence of predicting ‘more humid tomorrow’ for each case and construct an ROC curve for each classifier. You should be able to plot all the curves on the same axis. Use a different colour for each classifier. Calculate the AUC for each classifier.

7.            Create a table comparing the results in Questions 5 and 6 for all classifiers. Is there a single “best” classifier?

Investigative Tasks (18 Marks)

8.            Examining each of the models, determine the most important variables in predicting whether it will be more humid tomorrow or not. Which variables could be omitted from the data with very little effect on performance? Give reasons. (2 Marks)

9.            Starting with one of the classifiers you created in Question 4, create a classifier that is simple enough for a person to be able to classify whether it will be more humid tomorrow or not by hand. Describe your model, either with a diagram or written explanation. What  factors were important in your decision? State why you chose the attributes you used. Using the test data created in Question 3, evaluate model performance using the measures you calculated for Questions 5 and 6. How does it compare to those in Question 4? (4 Marks)

10.          Create the best tree-based classifier you can. You may do this by adjusting the parameters, and/or cross-validation of the basic models in Question 4. Show that your     model is better than the others using the measures you calculated for Questions 5 and 6. Describe how you created your improved model, and why you chose that model. What    factors were important in your decision? State why you chose the attributes you used. (4 Marks)

11.          Using the insights from your analysis so far, implement an Artificial Neural Network classifier and report its performance. Comment on attributes used and your data pre-processing required. How does this classifier compare with the others? Can you give any reasons? (4 Marks)

12.          Fit a new classifier to the data, test and report its performance in the same way as for previous models. You can choose a new type of classifier not covered in the course, or a new version of any of the classifiers we have studied. Either way, you will be implementing a new R package. As a starting point, you might refer to James et al. (2021), or look online. When writing up, state the new classifier and package used. Include a web link to the package details. Give a brief description of the model type and how it works. Comment on the performance of your new model. (4 Marks)

Report and Video Presentation (7 Marks)

Write a brief report (suggested length 8 – 10 pages) summarizing your results. Use commenting in your R script, where appropriate, to help a reader understand your code. Alternatively combine working, comments and reporting in R Markdown. (3 Marks)

Record a short presentation using your smart phone, Zoom, or similar method. Your   presentation should be approximately 5 minutes in length and summarise your main  findings, as well as describing how you conducted your research and any assumptions made. Pay particular emphasis to your results for the investigative tasks. (Submission Hurdle and 4 Marks)