EDA with R: From Basic Stats to Advanced Plots

Key Takeaways

Welcome. Let's say you're a data analyst at an important hospital and you're trying to understand all the different metrics that are associated with your patients and see how we can provide more care or actually better care. We'll be using the tidyverse library to achieve this goal. We'll be performing EDA or exploratory data analysis where we go into the data, try to understand if there's any issues and try to make a story out of it. First, let's begin with loading our library and our data set. All right, let's try to look at it. We see very common columns such as ids, age, other numeric metrics, some boolean and categorical metrics such as insurance, hospital visits, and diabetes status. All right, let's clean this variable. I'm sorry, this terminal. And let's start by calculating the mean for these numerical categories. Again, you can do mean or median. You want to make sure to add this parameter NA. RM equals true to avoid using any NAS in your analysis. Let's run this and see what happens. Beautiful. As we expected, we can see that we have created the mean values for all these categories. Age, PMI, blood pressure systolic, and blood pressure diastolic. Now, let's move on to the plot, the fun part. First, let's create a histogram using the GM histogram with a bin width of five, which is a customizable parameter. Let's run this. As you can see, we have a simple histogram but powerful. The insights we can graph from here is that the majority of our patients are in the 50 to around 60 age group. All right. Now, let's move in and add a new layer coloring. in this case by insurance type. We do this by using the fill parameter and adding the column that we're [clears throat] interested in as a color. Now, as you can see, we have the same plot, but now with extra coloring. This will allow us to see a big picture of the types of insurance that we have in our patient pool. In this case, public and private are the majority with very few having none. Now let's move into a different type of plot, a box plot. In this case, denoted by the geomeore box plot. We're also adding a new layer which is the labs parameter where we add titles. In this case, a nice overarching title BM BMI by gender as well as titles for the X and Y axis. Turn this As you can see, this bar is immediate. We also can see any outliers from this as well as our titles for our plots. Beautiful. This is great. All right. Now, we're going into a different type of plot, a scatter plot or a GM underscore plot. We also have this alpha parameter, which is the thickness of the points. Let's plot BMI versus blood pressure systolic and see what it looks like. All right. In this case, we can see a scattering, a lot of scattering, not much of a trend, but at least we can visualize everything. This will allow us to understand whether we have patients with really high blood pressure. In this case, for example, this one, very low BMI, but really high blood pressure. Might be worth looking into this patient. Now let's see as we're trying to understand the data we're trying to put numerical values to it. In this case correlation let's see if we can plot the correlation value for BMI versus blood pressure systolic which we plotted in the previous plot. Remember to use use equals complete.obs to only use complete observations or in other words entries that have no NAS. Let's clean the terminal so it's much cleaner. And this, as you can see, we have a correlation of nearly 1%. Quite low. It makes sense. A lot of the data was really scattered. But remember, you can do this with any numerical combination, so it's worth investigating. Now, let's circle back to the box plot and see if we can combine what we learned where we're adding a plot and then using the fill parameter to add some coloring to it. In this case, the insurance type versus the number of hospital visits. All right, let's see how this looks. Beautiful. Just as we expected, a box plot with coloring. This is fantastic. Remember all these plots are customizable. So it's at your disposal to try to understand what the best combination of color and viewing will attract your audience attention. Now let's circle back and try to understand what we did today and understand that we did a lot of parts of EDA starting with analyzing general statistics and then moving on to visualizations and customization of these visualizations where we learned about histogram plots, box plots and scatter plots. We also learned how to calcate a correlation which is a very powerful statistic. And there you have it. We learned a lot but you're doing great. So keep going and happy coding.