$9$:$56$

$.$ II $56$ :

DATA$\_514\_$Unit$\_3\_$F$24$

Task $3$:

Based on the information from the required reading assignments from Kuhn & Johnson, as well as based on what you've learned from Task $2,$ perform the following:

a$)$ Prepare Grant data for Tasks $3$b$)$ and $3$c$)$ experiments:

To prepare Grant data for these experiments, you need file unimelb$\_$training.csv and script CreateGrantData.R:

save unimelb$\_$training.csv in your RStudio working directory $($I have uploaded the file to Blackboard, but you could also find it on GitHub$).$

install the AppliedPredictiveModeling package and run the scriptLocation$()$ command to locate CreateGrantData.R; then update $($see below$)$ and run this script in your RStudio environment. ${?}^{11}$ Since the script is old and since R environment has changed, you need to make the following updates to the script:

add stringsAsFactors $=$ TRUE to the read.csv$()$ command,

add options$($expressions $=15000)$ near the beginning of the script. ${?}^{12}$

${?}^{11}$ In my experiments with this script, parallel processing did not significantly decrease the running time, so to turn it off, you may change one line at the beginning of the script: from "cores to "cores However, if you want to run it in parallel, you need to use doParallel$()$ instead of doMC$().$

${?}^{12}$ Extra credit is available for a good explanation of why these options are needed to run the script under R $4.$x$.$x Page $19$ of $20$

page $20$

b$)$ Using Grant data perform LDA experiments; build and test an LDA classification model.

c$)$ Using Grant data perform Partial Least Squares Discriminant Analysis $($PLS$-$DA$)$ experiments: using the function identify a PLS$-$DA model with optimal number of PLS components, and then test this classification model.

Important instructions for your report:

For Tasks $2$ and $3$ experiments, provide description of all steps of your experiments, their results $($including confusion matrix and at least the basic performance metrics: accuracy, sensitivity$,$ and specificity; use the confusionMatrix$($ function from the caret package$),$ and discussion of the results. Include $-$ within your narrative for each step $-$ the R code used at this step, as well as printouts of its most important results. The R code included at each step must be complete, that is$,$ when copied from your report and executed, it has to work.

Note: Set seed to $100$ before any command that uses the random number generator $($RNG$),$ so your results are the same as expected.

Page limit: max $12$ pages $($plus references$).$

Very important: Make sure that you follow all report requirements as specified in Syllabus.