What Is Data Science?

Data science is an interdisciplinary field focused on extracting meaningful information from large sets of data. To discover hidden patterns, Data Scientists use math, science, algorithms, and systems to identify opportunities for increased efficiency, productivity, and profitability.

In simpler terms, data science uses math and technology to find hidden patterns (and ways to be more productive and profitable) in raw data. To find those patterns, a Data Scientist spends a lot of time collecting, cleaning, modeling, and examining data, from numerous angles, some of which have not been looked at before.

Essentially, data science is about knowledge creation: it makes use of the most state-of-the-art techniques and tools the fields of computer science and statistics have to offer to turn a mess of data into knowledge that an organization can use to inform their business practices.

Among the most noteworthy techniques a Data Scientist uses are predictive causal analytics, prescriptive analytics, and machine learning. The first, predictive causal analytics, uses data to predict the likelihood of different possible outcomes of a future event. Prescriptive analytics goes a step further, suggesting a range of different actions based on those possibilities, with an eye toward optimizing outcomes. Machine learning, unlike the two techniques just mentioned, is not the “what” but the “how” of data science: it’s the practice of using data-based algorithms that improve automatically based on past experiences – essentially learning to do their job better – to discover patterns and make predictions.

That said, in the real world, the practice of data science involves much more than simply using computers to crunch numbers. In fact, Data Scientists may be heavily involved in the decision-making process across departments, which means that, practically speaking, data science also involves collaborating with others, and especially knowing how to communicate important findings to other people.

What Do Data Scientists Do?

The common perception that Data Scientists crunch numbers is not too far off the mark; they do work with large sets of data, deciding what data is needed, cleaning the data, building models of what the data can show, and organizing it to reveal latent information—and this effort is always directed toward some kind of goal.

Notably, those data sets aren’t always numbers. While most Data Scientists do work with numerical data (73 percent, according to the BrainStation Digital Skills Survey), there are other types of data as well. According to the same survey, 61 percent of respondents work with text, 44 percent with structured data, 13 percent with images, and 12 percent with graphics—even video and audio are ripe for analysis, with 6 and 4 percent (respectively) of respondents working with these media regularly.

These results hint at the way data science is expanding far beyond the world of financial tables, and exerting its influence in areas like maximizing customer satisfaction and extracting valuable insights from social media.

As a result, every industry has its own types of data, and its own ways to leverage that data to help meet desired outcomes. In every case, though, data science serves as a way to help leadership make better, more informed decisions—whether that’s improving a product, understanding a new market, retaining customers, effectively deploying a labor force, or making better hires.

Data Scientists, therefore, use a combination of techniques and concepts, including:

Descriptive Analytics

Studies large sets of data to understand the way things are, including correlations and even causations that aren’t immediately obvious.

Predictive Causal Analytics

Draws inferences from data using a variety of statistical techniques—including data mining, predictive modeling, and machine learning—to predict the possibilities of a future event.

Prescriptive Analytics

Provides intelligence-based recommendations to produce a desired outcome or accelerate the results of a given application or business process.

Machine Learning

To put it simply, machine learning – or the process of a computer learning how to better perform a task as it gains more experience doing so – uses algorithms to make predictions and find patterns. Machine learning spans a wide array of ideas, tools, and techniques used by Data Scientists and other professionals, and it’s one of the most popular methods for processing big amounts of raw data.

It might be easiest to view machine learning as a part of data science. Machine learning frees Data Scientists from the tedious task of sifting through massive volumes of data by using complex algorithms and problem-solving methods including supervised and unsupervised learning, regression, classification, clustering, and neural networks.

Examples of machine learning are all around you. Facebook, for instance, uses machine learning to analyze your past behavior to present content and notifications in line with your interests. Similarly, when Netflix somehow recommends a show you’d love to binge-watch, it’s an example of machine learning.

Perhaps the simplest example of machine learning in motion lies in how it approaches the task of recognizing handwriting. To train a machine with examples of correct input-output pairs – which is called supervised machine learning – the computer is shown images of handwritten numbers alongside the correct labels for those digits. The computer then tries to figure out the shared characteristics of each digit, and gradually picks up on the patterns between the images and the labels.

Generally, machine learning is effective to solve problems that are statistical or probabilistic in nature, deeply complex, and that can be adequately handled with an approximate solution. For instance, the issue of detecting credit card fraud checks those boxes: solutions are probabilistic because a determination won’t be made until a company reaches its customer; the rules around fraud are complex; and approximate solutions are adequate since we’re simply flagging transactions for further review.

Although many of the more advanced machine learning tools do require some experience and knowhow, the basics can still be impactful for those looking to dig deeper. Many supervised and unsupervised learning models are implemented in R and Python, and straightforward models like linear or logistic regression can be used to perform informative machine-learning tasks.