Numeric variables can be considered to be either continuous or discrete. The reason why this may be significant is that certain machine learning algorithms and analysis approaches do not work well with continuous variables, and so in some cases, we may have to adjust the data that is stored in order to optimize it for consumption by those types of algorithms. A continuous variable is a variable that is quantitative and that is going to represent an uncountable number of values. In fact, it could extend infinitely. For an example, if we looked at my white board here, you'll see that if we were to work with a continuous variable, that perhaps we have a variable that is defined where we're calling it time. Now how would we represent time? Perhaps by a timeline. I'm going to draw time here and we're going to have a starting time perhaps and ending time. Now, when we look at the starting time and the ending time here, you'll see that if we were to try to measure all of the possible values between 12 o'clock and 12:09, the range of values included here would actually include all real values between those two points in time, and so because of this, there literally is an infinite number of values between those two points. This is because we could continue to divide the time up into smaller and smaller and smaller units of time down to seconds, milliseconds, microseconds, and so on. When we simply say that we're going to record a value like time, we must understand that that is a continuous variable and as such, it may not be useful for certain types of algorithms. Now, on the other hand, we may choose to convert our time into a discrete variable. When we're talking about a discrete variable, we have a finite countable number of values, and so as we're looking at this, we might define a variable that now might be something like time in minutes. Now by providing that context where we're now going to be clustering our times only by the minute we are going to create a fixed number of possible values. Again, if we examine our timeline and we were just start at 12 o'clock and go through to perhaps 12:09, at this point now, we would be able to say, okay, we have 1,2,3,4,5,6,7,8,9,10 possible values that could exist on that timeline, and because there are only 10 possibilities, then of course, this limits the number of values that are going to be presented through to our algorithm. One example of an algorithm that does not handle continuous variables very well would be a tree algorithm like CART, and because the tree would keep splitting and splitting and splitting on all of those unique little variations in value. Whereas if we could adjust those variables to be discrete variables then the number of splits would be reduced down to a much lower value creating a much simpler tree and one that of course, would not overfit the data as well. This is an example of why we may want to consider using discrete variables for certain cases over continuous variables. Of course, in some cases, using the continuous variables would be appropriate because you may need to maintain that level of precision.
