Data Visualization

Signal Processing

Signal processing is used in applications such as speech analysis, heart rate monitoring, wireless communications, remote sensing, climate monitoring, and GPS. Common tasks involve preprocessing and comparing signals, designing digital filters, transforming signals, performing measurements, and detecting patterns and events. Data visualizations are used to analyze signals of interest in time, frequency, and time-frequency domains.

The plot below is the audio data from a Pacific blue whale. The visualization is created in MATLAB^® using the Signal Analyzer app that helps to visualize signals in the time and frequency domain.

Image Processing and Computer Vision

Image and video processing helps to find shapes, count objects, identify colors, measure object properties, and find other meaningful information. Image processing techniques are often applied as a preprocessing step in the computer vision workflow. Applications in this domain include facial recognition for smart phones, pedestrian and vehicle avoidance in self-driving vehicles, video surveillance, tumor detection in medical MRIs, and other image retrieval systems.

For example, BMW uses computer vision capabilities in the Assisted Driving View (ADV) to depict surrounding vehicles and identify their types.

Artificial Intelligence (AI)

Data visualization plays an important part in developing AI models—using machine learning or deep learning—because the models rely on large data sets, which are difficult to interpret. In machine learning, cluster analysis helps in detecting anomalies and data preprocessing in supervised learning. Principal component analysis (PCA) and t-distributed Stochastic Neighbor Embedding (t-SNE) are the two most used data visualization techniques because they help reduce data dimensions so you can focus on key differentiating dimensions.

In deep learning, you can monitor training progress using data visualizations like plots of network accuracy and loss and investigate trained networks using visualization techniques such as gradient-weighted class activation mapping (Grad-CAM), occlusion sensitivity, local interpretable model-agnostic explanations (LIME), and deep dream.

How Does Data Visualization Work?

Software packages provide capabilities to transform raw data into rich visualizations, such as plots, charts, and diagrams. The following is an illustrative example of bike traffic density data. Solely by visually inspecting the raw data, it is difficult to establish a relationship between the data points.

The bar graph below shows that the bicycle traffic density rises and falls over the days of week. It’s now clear that the number of bicyclists is greater during the weekdays as compared to weekends. This visualization allows us to infer that bicyclists on this route are predominantly commuting back and forth from work.

A scatter plot can be used to get more insights from the same data. The following plot shows the total number of bikes heading eastbound and westbound at specific times of the day. Based on this plot, we can conclude that eastbound routes lead to business districts and westbound routes lead to residential areas. Furthermore, we can establish that rush hour traffic is from 8:00—10:00 a.m. on the eastbound route and 4:00—6:00 p.m. on the westbound route.

A swarm chart is a special kind of scatter plot that can show density patterns of bike traffic for different times of the day, day of the week, and direction.

In the bike traffic example, visualizing data with different types of plots like bar, scatter, and swarm charts helps us extract useful information from the data set including peak traffic days, direction of commute, and busiest time of the day.