During the master's degree, I designed an interactive data visualization to make visually accessible the data regarding data breaches that have occurred in the last 18 years.

Introduction

The purpose of the project was to create an interactive data vis in Tableau starting from a data set of our choice. Once created, the visualization had to be evaluated with the target users in order to find problems during the interaction and fix them.

Project output

Starting from these premises, I designed and developed a desktop dashboard made up of multiple graphs arranged on a vertical page.

The data vis represents a list of data breaches, using data coming from various sources, that involve the number of stolen records, the methods employed, the locations and type of the organizations that were affected by the breaches, in a time period that spans from 2004 to 2021.

The interactive version of the dashboard is available here.

Dataset description

The dataset that I used was available on Kaggle, however the original source of data came from Wikipedia. This dataset contained all the main data breaches that have occurred in the world from 2004 until 2021. The dataset had 293 rows, which is the number of entries of the reported data breaches.

From the Wikipedia article, it has been reported that the data was collected from various sources, including press reports, government press releases and news articles. The list consisted of all data breaches that resulted in theft of 30,000 or more records, even though many minor breaches still occur continuously.

The dataset contained 5 variables:

• Entity.
  The name of the company, organization or institution that has undergone an attack
• Year.
  The year in which the data breach occurred
• Records.
  The number of records that have been stolen
• Type of organization.
  The field to which the attacked organizations belong to
• Method.
  The cause that led to the data breach

Data processing

The dataset from Kaggle did not contain any missing data.

However, I made some adjustments to the type of organization variable. Originally, the number of categories contained in the column was 54, but they were often very similar to each other. In addition to this, some organization types were reported with semantic errors. At the end of this process, the number of categories was reduced to 19.

Then, I decided to add another variable to the dataset in order to better understand the geographical impact of the phenomenon. So, I added the “State of entity” column, which contains the countries where the attacked organizations have their registered offices.

Target users

My visualization aims at different types of users: from those who have an unclear idea on the subject and want to approach the world of IT security, up to those who have a good understanding on the subject but want to use the dashboard to deepen their knowledge thanks to the interaction with the infographic. In general, I address to young people who frequently uses digital devices, without being fully aware of the risks.

Furthermore, given the structure and interactivity of the dashboard, it can only be used on a computer's desktop, so I expect that the target users are already familiar with it.

Graphs architecture

Starting from the updated dataset, I created in Tableau the most relevant graphs to communicate the phenomenon, trying to give a complete overview through different types of representation. In particular, I designed:

• A bar chart.
  I decided to show the total amount of data stolen in the different organization types as the first visualization.
  The bar graph allowed me to clearly show the quantitative differences between organizations and the absolute number of stolen records organized in descending order.
• A word cloud chart.
  The second visualization was related to the methodologies used to carry out data breaches.
  I have decided to use a word cloud, in which the amount of stolen data is expressed hierarchically both by the size of the word indicating the causes of breaches and by the color value assumed by the word.
  Since it has a limited number of methods, this view is the most effective from an informative and perceptive point of view.
• A line chart.
  The third visualization shows the amount of data stolen during the time period that ranges from 2004 to 2021.
  Since the time of the dataset is expressed in a discrete way, I have opted for the use of a line graph, which allows me to clearly show the trend of the phenomenon over the years.
• A bubble chart.
  The fourth view shows the number of data stolen from various companies in the dataset.
  This graph seems relevant to me because it allows users to recognize the names of well-known international companies to which they are used to entrust their data. The use of a bubble chart allows on one hand to quickly visualize the most affected companies and on the other, to outline the extent to which companies were involved in the attacks.
• A map chart.
  The main purpose of this view is to highlight the territorial distribution of the attacks in a impactful and clear way.
  In this case the amount of stolen data is expressed by the color value of the state, in relation to the legend on the bottom right of the graph.

Organization

The dashboard layout was created in Figma and added in Tableau.
At the top of the dashboard I provided some information to the visualization to allow users to better understand the structure and the interactive use of the page.
In particular, the first paragraph provides users with information about the content of the data visualization, in order to introduce them to the topic. The second paragraph contains the instructions to the users on how to discover the interactive nature of the dashboard.
Different graphs have been created below the introductory part. Each graph comes with a title and a description, which clarify the meaning of the visualization. The sections containing the interactive graphs come with a graphic symbol which allows the graphs to be recognized rather than remembered.
The division of the dashboard into sections has been designed to clearly distinguish each view from the others.

Visual choices

For my project, I used a dark neuromorphic design. It maintains a clean and modern visual communication and it ensures good internal coherence between graph sections.

The decision to use red monochromatic scales, representing quantitative data, on a dark background, has allowed me to obtain an impactful overview of the graphs, which seeks to evoke the seriousness of the topic while still being easy to read.

For the texts, I used the Roboto typeface family for the titles and Mukta for the paragraphs. I chose Roboto for its mechanical skeleton and its geometric forms and Mukta for its contemporary style.

Interactivity

The dashboard guarantees some interactivity to the users. In fact, it is possible to perform some filtering operations on two specific graphs of the visualization and observe the changes in all the other graphs at the same time. Moreover, I have provided not only for the possibility to interact with a single graph, but also for the possibility to select elements of the two graphs in succession to perform two consecutive filtering operations and to obtain a more detailed selection.

I have chosen two graphs to be interactive: one related to attacks on organizations and the other related to the hacking methods used. Both have been chosen because they allow the user to clearly view and select directly from the graph all the categories present in the dataset for that specific section. In addition to this, they are both data-rich visualizations on which it makes sense to perform filtering operations to carry out more specific analyses.

User testing

In this phase, I observed 5 users perform three specific tasks that required them to interact with the views, in order to detect the biggest errors committed.

The tested users were between 18 and 35 years of age and they had average digital skills. Four out of six users were already familiar with the topic of the dashboard. For each task I observed the execution time on the visualization, the effective achievement of the tasks, the number and the type of committed mistakes. The three tasks to be performed were:

• Task 1. Observe the dashboard and indicate which is the most used hacking method, then indicate the country in which most records were stolen.
  
  The goal of the first task was related to verifying the ease of reading and comprehension of the dashboard.
  



• Task 2. Select a hacking method of your choice and tell me the names of the first three companies to which most records were stolen. After that, return to the initial state.
  
  The goal of the second task was related to verifying the effectiveness of the single filtering method and the comprehension of the filtered information.



• Task 3. Comment on what happened over the years to the organizations of financial type hacked due to poor security.
  
  The goal of the third task was related to verifying the effectiveness of the double filtering method and the comprehension of the filtered information.
  

Error analysis

The user testing revealed some issues mainly in the accomplishment of the first task, in which the user had to become familiar with the Tableau environment and with the dashboard. This is highlighted also by the execution times, which were longer for the first task, even though it was the easiest one. Once the users understood the general functioning of the dashboard, on average they accomplished the next tasks without any problems.

Here I display the errors made by the users and their frequencies, with the aim to propose some revisions to the dashboard for the most important mistakes:

• Task 1.
  
  3/6 users did not correctly identify the most used hacking type (which was "hacked"). One user stated that he did not indicate “hacked” as an attack method because he thought it was an umbrella term for other words, rather than a methodology.
  
  1/6 users users did not figure out from the introduction the meaning of a "data breach", because he was not familiar with the topic.



• Task 3.
  
  3/6 users did not immediately comprehend the graphs which could filter the information. In fact, he attempted to click also on the graphs that did not allow it.
  
  2/6 users did not correctly interpret the information of the line chart. In fact, they said that the graph showed the number of attacks over the years, rather than the number of stolen records over the years.

Psychometric questionnaire

During the evaluation phase I sent the short version of the User Experience Questionnaire (UEQ) to a sample of 19 users to statistically evaluate some aspects of the dashboard.

The scales of the questionnaire cover a comprehensive impression of user behavior. Both classical usability aspects (efficiency, perspicuity, dependability) and user experience aspects (originality, stimulation) are measured.

The questionnaire was sent to a sample of 19 users via Google Forms. They filled it in without supervision. The sample users were between the ages of 20 and 35, with average digital skills and with a middle-high level of education.

Questionnaire results

From the questionnaire results, I noticed that in general the dashboard achieved positive results, with some differences between pragmatic quality and hedonic quality.

In the results table, mean values between -0.8 and 0.8 represent a neutral evaluation of the corresponding scale, values > 0.8 represent a positive evaluation and values < -0.8 represent a negative evaluation. The range of the scales is between -3 (horribly bad) and +3 (extremely good). But in real applications, only values in a restricted range will be observed.

As shown in the table and in the graphs below, in my case it is possible to observe higher values for hedonic quality items than for pragmatic quality ones. In particular, the sample users assigned higher grades for the inventiveness and the original style of the dashboard. It is possible to notice slightly lower grades, but still positive ones, to the ease of use of the dashboard in the pragmatic quality entry.

Then I compared the results for the evaluated product with the data related to benchmark categories. The measured scale means are set in relation to existing values from a benchmark data set. The data set that I have used contains data from 21175 persons from 468 studies concerning different products.

The graph below shows that the dashboard obtained results all above the average and it confirms that the hedonic quality is the scale that obtained the best results. In particular, the visualization is in the 10% of the best results obtained in the evaluation of the products referred to the benchmark dataset.