• My app aims at exploring some more machine learning, at least conceptually.
• Among the various stock datasets that R provides, there is one called airquality, containing some information about pollution.

  Ozone Solar.R Wind Temp Month Day
1    41     190  7.4   67     5   1
2    36     118  8.0   72     5   2
3    12     149 12.6   74     5   3
4    18     313 11.5   62     5   4
5    NA      NA 14.3   56     5   5
6    28      NA 14.9   66     5   6

• The database, already small as it is, (153 observations of 6 variables), is even thinner if one excludes incomplete cases from the analysis.
• I picked a small one because of the heaviness of the training step that comes within a machine learning algorithm.
• The incomplete cases are overall the 27.5% of the total
• They can be dealt it in two ways: either imputing them with the median or eliminating them altogether from the dataset. That’s what the “Impute NAs” checkbox is for.

The app also lets the user choose the data partition (in a limited range) with the “Training partition” slider; this will set how much of the data will be used for training.
The training takes place using a plain simple linear model. With a k-fold cross-validation training control system.

• The app also lets the user choose how many (in a range 2 to 5) folds does he/she want to perform cross-validation with; that’s what the last input is for.

• Please note that above all this, a preliminary analysis on the relevance of the 5 predictors has been made, and it turned out all predictors were somewhat relevant.

• FINAL REMARK: this is NOT (nor it claims to be) a good exercise on machine learning. It was just something to recall a few concepts with, while experimenting a little with the tools of this course.