The human mind is often good at seeing patterns, trends, and outliers in visual representations. The large amount of data present in many data science problems can be analyzed using visualization techniques. Visualization is appropriate for a wide range of audiences, ranging from analysts, to upper-level management, to clientele. In this chapter, we present various visualization techniques and demonstrate how they are supported in Java.
In this chapter, we will illustrate how to create different types of graph, plot, and chart. The majority of the examples use JavaFX, with a few using a free library called GRAphing Library (GRAL). There are several open source Java plotting libraries available. A brief comparison of several of these libraries can be found at https://github.com/eseifert/gral/wiki/comparison. We chose JavaFX because it is packaged as part of Java SE.
GRAL is used to illustrate plots that are not as easily created using JavaFX. GRAL is a free Java library useful for creating a variety of charts and graphs. This graphing library provides flexibility in types of plots, axis formatting, and export options. GRAL resources (http://trac.erichseifert.de/gral/) include example code and helpful how to sections.
Visualization is an important step in data analysis because it allows us to conceive of large datasets in practical and meaningful ways. We can look at small datasets of values and perhaps draw conclusions from the patterns we see, but this is an overwhelming and unreliable process. Using visualization tools helps us identify potential problems or unexpected data results, as well as construct meaningful interpretations of good data.
One example of the usefulness of data visualization comes with the presence of outliers. Visualizing data allows us to quickly see data results significantly outside of our expectations, and we can choose how to modify the data to build a clean and usable dataset. This process allows us to see errors quickly and deal with them before they become a problem later on. Additionally, visualization allows us to easily classify information and help analysts organize their inquiries in a manner best suited to their particular dataset.
There are many types of visual expression available to aid in visualization. We are going to briefly discuss the most common and useful ones, and then demonstrate several Java techniques for achieving these types of expression. The choice of graph, or other visualization tool will depend upon the dataset and application needs and constraints.
A bar chart is a very common technique for displaying relationships in data. In this type of graph, data is represented in either vertical or horizontal bars placed along an X and Y axis. The data is scaled so the values represented by each bar can be compared to one another. The following is a simple example of a bar chart we will create in the Using country as the category section:
A pie chart is most useful when you want to demonstrate a value in relation to a larger set. Think of this as a way to visualize how large the piece of pie is in relation to the entire pie. The following is a simple example of a pie chart showing the distribution of population for selected European countries:
Time series graphs are a special type of graph used for displaying time-related values. These are most appropriate when the data analysis requires an understanding of how data changes over a period of time. In these graphs, the vertical axis corresponds to the values and the horizontal axis corresponds to particular points in time. In particular, this type of graph can be useful for identifying trends across time,or suggesting correlations between data values and particular events within a given time period.
For example, stock prices and home prices will change, but their rate of change varies. Pollution levels and crime rates also change over time. There are several techniques that visualize this type of data. Often, specific values are not as important as their trend over time.
An index chart is also called a line chart. Line charts use the X and Y axis to plot points on a grid. They can be used to represent time series data. These points are connected by lines, and these lines are used to compare values of multiple data at one time. This comparison is usually achieved by plotting independent variables, such as time, along the X axis, and independent variables, such as frequency or percentages, along the Y axis.
The following is a simple example of an index chart showing the distribution of population for selected European countries:
When we wish to arrange larger amounts of data in a compact and useful manner, we may opt for a stem and leaf plot. This type of visual expression allows you to demonstrate the correlation of one value to many values in a readable manner. The stem refers to a data value, and the leaves are the corresponding data points. One common example of this is a train timetable. In the following table, the departure times for a train are listed:
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
However, this table can be hard to read. Instead, in the following partial stem and leaf plot, the stem represents the hours at which a train may depart, while the leaves represent the minutes within each hour:
|
Hour |
Minute |
|
06 |
:15 :20 :25 :30 :40 :45 :55 |
|
07 |
:15 :20 :25 :30 :40 :45 :55 |
|
08 |
:00 :12 :24 :36 :48 |
|
09 |
:00 :12 :24 :36 :48 |
|
10 |
:00 :12 :24 :36 :48 |
This is much easier to read and process.
A very popular form of visualization in statistical analysis is the histogram. Histograms allow you to display frequencies within data using bars, similar to a bar chart. The main difference is that histograms are used to identify frequencies and trends within a dataset while bar charts are used to compare specific data values within a dataset. The following is an example of a histogram we will create in the Creating histograms section:
A scatter plot is simply collections of points, and analysis techniques, such as correlation or regression, can be used to identify trends within these types of graph. In the following scatter chart, as developed in Creating scatter charts, the population along the X axis is plotted against the decade along the Y axis:
Each type of visual expression lends itself to different types of data and data analysis purposes. One common purpose of data analysis is data classification. This involves determining which subset within a dataset a particular data value belongs to. This process may occur early in the data analysis process because breaking data apart into manageable and related pieces simplifies the analysis process. Often, classification is not the end goal but rather an important intermediary step before further analysis can be undertaken.
Regression analysis is a complex and important form of data analysis. It involves studying relationships between independent and dependent variables, as well as multiple independent variables. This type of statistical analysis allows the analyst to identify ranges of acceptable or expected values and determine how individual values may fit into a larger dataset. Regression analysis is a significant part of machine learning, and we will discuss it in more detail in Chapter 5, Statisitcal Data Analysis Techniques.
Clustering allows us to identify groups of data points within a particular set or class. While classification sorts data into similar types of datasets, clustering is concerned with the data within the set. For example, we may have a large dataset containing all feline species in the world, in the family Felidae. We could then classify these cats into two groups, Pantherinae (containing most larger cats) and Felinae (all other cats). Clustering would involve grouping subsets of similar cats within one of these classifications. For example, all tigers could be a cluster within the Pantherinae group.
Sometimes, our data analysis requires that we extract specific types of information from our dataset. The process of selecting the data to extract is known as attribute selection or feature selection. This process helps analysts simplify the data models and allows us to overcome issues with redundant or irrelevant information within our dataset.
With this introduction to basic plot and chart types, we will discuss Java support for creating these plots and charts.