A Beginner’s Guide to Data Mining Techniques

Inna Logunova
Article by Inna Logunova
June 9th, 2022
12 min read

Data mining techniques have applications in all areas from business to science and governance. Companies use data mining to analyze recorded data, such as user preferences, sales figures, and historical inventory levels. If they are able to identify trends and recurrent patterns in this data, they can make better decisions. When managed properly, this information can become an effective tool to drive brand awareness, product development, and marketing initiatives, and strengthen an overall business development strategy.

In this blog post, we will look at how data mining differs from machine learning and what data mining techniques can be used to turn raw data into business insights.

What is data mining?

Data mining (DM) is a computer-assisted process to find patterns in big datasets. Data mining applies intricate algorithms to bring them to the surface so they could be used for solving real-world problems.

Although there are several types of data mining, they usually fall into two general categories: exploratory and predictive.

Exploratory and predictive data mining

Exploratory data mining has been known for more than 50 years. In the past century, it was widely used in statistics to determine the applicability of certain techniques for data analysis. In practical terms, it could be a tool to detect fraudulent insurance claims, such as repeated photographs of damaged goods submitted for multiple insurance cases. Another example is highlighting incorrect sampling – for instance, where 90% of respondents were women instead of the required 50%. In general, exploratory data analysis (EDA) describes data distribution, helping identify anomalies or verify hypotheses based on the graphical or non-graphical presentation of big data. Find out more about EDA in this post.

Predictive data mining is a 21st-century technology that has been around for two decades. The field evolved from the 1980s artificial intelligence research that focused on how computers can learn from large amounts of unspecified data. To stick with the example of an insurance company: by feeding all records (policy numbers, addresses, etc.) into an algorithm, you can detect specific patterns, like the anomaly high number of claims from a particular organization or persona or irregularities in specific cases. Thus, the irregularities in policy prolongation can be a signal of a low level of customer satisfaction.

To clarify the difference between exploratory data analysis and predictive data mining, we can add that the first term refers to the process of the general evaluation of raw data at a more abstract level. It is used to examine whether the collected data has some anomalies or discrepancies and whether it conforms to normal distribution or another distribution law. This can help avoid working with incomplete data samples or statistical methods only applicable to normal distribution arrays.

In predictive DM, the goal is to uncover non-obvious, multi-factor correlations between figures, especially where statistical methods are not applicable.

Not to be confused: data mining vs. KDD

Data mining is a part of the procedure referred to as Knowledge Discovery in Databases (KDD). That often creates confusion between the two, as a number of sources use these concepts interchangeably.

Whereas KDD is a general process of extracting knowledge from data, data mining is a stage within KDD that deals specifically with the recognition of patterns in data. In other words, data mining is the application of a particular algorithm for the overall purpose of the KDD process.

KDD is iterative, and during this process, various adjustments can be made, including the refinement of evaluation and mining and adding new data that will help obtain better results.

To better understand the difference between data mining and KDD, you can watch the video below.

Where is data mining (DM) used?

It is widely used in a variety of industries, including healthcare, retail, finance, government, and manufacturing.

For example, if a company wants to discover patterns or trends among customers that buy certain products, it can use data mining techniques to analyze their purchasing history and develop models that predict which customers wish to buy specific goods based on their demographics or behavior. Thus, in retail, data mining helps companies develop more successful sales strategies.

In addition, these tools can be used to:

  • Segment customers: identify groups of customers that share similar behaviors and target them with personalized marketing messages.
  • Predict cancelations: find out which customers tend to cancel their orders based on historical data.
  • Detect fraud: based on historical transaction data, it is possible to identify suspicious behavior and block it.
  • Recommend products and services to users depending on their past experience.

Examples in other areas

Data mining techniques are also gaining ground in education, science, logistics, finance, and banking – in other words, virtually every sphere.

data mining technique applications

In education, DM helps build customized programs based on:

  • Students’ learning patterns – for instance, their tendencies to consume information through video audio, text, or a combination of the three.
  • Labor market trends – this allows to determine the most relevant educational focus.

In finance, data mining is used to:

  • identify investment opportunities;
  • predict demand for some stock shares, which enables potential investors to make informed decisions.

Data mining also has applications in law enforcement and intelligence:

  • Customs officers can better understand the typical profile of border violators based on border-crossing history and focus on specific categories of individuals.
  • Police can identify areas where they need to deploy more manpower, knowing when and where the likelihood of a crime is highest.

What is the difference between data mining and machine learning?

The concepts of data mining and machine learning are similar and therefore are often used interchangeably. Both analyze datasets to make predictions and gain insights. However, they are based on different principles. In ML, analysis is preceded by setting criteria for data categorization. Since this step foregoes data clearance, it allows to dismiss unsuitable data from analysis. In DM, patterns are not known beforehand and have to be established.

Data mining uses algorithms to discover correlations and interdependencies in data and decipher their meaning, for example, customer preferences. One example is the discovery of periodic orders of pet food or shampoo to remind customers and encourage them to buy from the company.

Take another example. When a trading company wants to place an order for production based on past sales, it needs to find the best combination of items taking into account several factors.

The order should:

  • satisfy the increasing demand for the best sellers;
  • predict the optimal production of new items;
  • take into account seasonal fluctuations;
  • compensate for the lack of out-of-stock units;
  • replace certain SKUs with similar goods;
  • optimize the stock so it remains within the available space and the agreed cash flow.

Mathematical methods can only solve part of the problem, while data mining can provide a better solution.

Machine learning is a subset of AI and is about designing algorithms that learn from data and improve with experience. A spam filter is a common example of machine learning. Algorithms analyze each email and look for patterns that indicate whether it is spam or not (e.g., containing the words “free money” or coming from a suspicious domain). Machine learning algorithms are often used in e-commerce platforms and streaming services like Amazon and Netflix to make product recommendations. They analyze customers’ previous purchases and search history to determine what they might be interested in buying next.

Machine learning algorithms can be used for clustering, classification, regression (predictive analysis), association rule development, and anomaly detection, meaning they are more universal in their application, helping find out general trends and patterns.

Data mining methods are used to work with customer data and recognize similarities in a particular segment. So, depending on the task, you can use ML or data mining. In many cases, they complement and enrich each other. For example, data mining can help establish hypotheses that will subsequently be used for machine learning. Also, ML techniques can be used to verify these hypotheses.

Data mining Machine learning
What it is the process of extracting information from data a subset of artificial intelligence that allows machines to make predictions based on data
How it works identifies rules and patterns in large amounts of data uses multiple approaches to train machines to learn without human intervention
Purpose establish data correlations and discover sequences and trends for generating hypotheses check hypotheses and evaluate their probability

Stages of the data mining process

Preparatory stage

Setting business goals

The first step is to determine the ultimate goal of the project and figure out how it will benefit the organization. The goal may be to better understand sales trends, classify consumers based on their preferences or behavior, or predict buying tendencies.

Data extracting and cleansing

The following stage is to collect relevant data from a variety of sources, such as CRMs, databases, web pages, social media, etc. You will need to merge data from all these channels and transform it into a format that can be used for research (analysis).

Once you have the data you need, you have to pre-process it so that it is ready for analysis. This involves data cleansing and structuring.

Check out this post about data preprocessing to get a better understanding of this stage.

Data mining proper

Data exploration

Before getting down to analyzing data, it is essential to understand it. The purpose of data exploration is to pinpoint patterns or correlations in data.

Hypotheses formation

After examining the data, it’s time to identify unknown clusters, patterns, or trends. During this stage, algorithms for classification, prediction, and clustering are applied. Each hypothesis is evaluated using appropriate techniques such as cross-validation, bootstrapping, and error matrix analysis. The most valuable hypotheses are accumulated and later presented to the public.

Post-processing: presentation

So that the results could turn into valuable business insights, they must be demonstrated in a clear, structured, and easily understandable form. Visualizing them as a report, diagram, or infographic is a way to highlight the most important discoveries such as trends, patterns, or correlations that will allow data-driven decision-making.

The graphic below summarizes all data mining stages.

data mining steps

Data mining techniques

To extract information from data, a wide variety of data mining techniques are employed.

These include:

  • classification
  • clustering
  • association rule learning
  • regression
  • anomaly detection
  • sequential pattern mining

Depending on data characteristics, batch or real-time processing can be used. The first one works for big amounts of data collected over a certain period. Real-time processing applies to systems with dynamically updated data, the example of which is Google Analytics real-time overview report that reflects the website user activity happening here and now.

Classification

Classification is used to divide data into predetermined groups or classes. This data mining technique determines the class to which a record belongs based on the values of several attributes. The goal is to sort data into predefined classes. Most commonly, classification involves predicting a target variable that can take on one of two or more possible values (e.g., spam/not spam; positive or neutral/negative review) given one or more input variables called predictors.

Classification shown

Take a few minutes to watch this video that explains how classification works on real data.

Clustering

Clustering is a technique for grouping related entries in a database into clusters based on their similarities. Whereas classification assigns variables into known categories, the clustering technique first singles out these clusters in the dataset and then groups variables based on their characteristics.

For example, you can cluster customers into groups according to the sales data – those who regularly buy pet food or specific drinks and who are stable in their preferences and customer behavior. Once you establish these clusters, you can easily target them with customized advertisements.

Clustering has a wide range of applications:

  • medical diagnostics
  • computational biology
  • text mining
  • web analytics

Association rule learning

Association rule learning discovers if-then patterns between two or more variables. The simplest example is the association between buying bread and butter. People that buy bread usually get butter with it, and vice versa. That is why you will find these two products close to one another in a grocery store.

However, the link may be not that direct. For instance, in 2004 Walmart discovered that the sales of Strawberry Pop-Tarts were at their peak before the hurricane. People stocked up not only the necessities like batteries but also these popular desserts. In retrospect, the psychological motivation is quite obvious: during emergencies, your favorite food gives you a sense of security, and tarts with a long shelf life are a perfect option. But to determine this relationship, it was necessary to apply data mining techniques.

Association rule learning

Regression

Regression establishes a relationship between variables. Its goal is to discover the right function that describes the relationship. If a linear function (y = ax + b) is used, the process is called linear regression analysis. For other types of dependencies, methods such as multiple linear regression, polynomial regression, etc. can be used.

Check out this overview of regression methods for more details.

Its most common application is planning and modeling. One example is forecasting customers’ age based on their purchase history. We can also predict costs based on such variables as consumer demand – for example, a surge of prices on the secondary market due to the increased demand for cars in the US.

Anomaly detection

Anomaly detection is a data mining technique used to identify outliers (values that deviate from the norm). For example, in e-commerce datasets, it can detect unusual sales during a given week at a store location. Among other things, it can be used to discover credit or debit fraud and identify intrusion or interruption in the network.

This video gives a simple explanation of outliers.

Sequential pattern mining

Sequential pattern mining is a data mining area that detects meaningful relationships between occurrences. Identifying a time-ordered sequence of events that happen with a specific frequency allows us to speak of a dependency between them.

For a more detailed overview, read this article.

Let’s say we want to investigate the impact of a medication or a particular therapeutic method on the life expectancy of cancer patients. Sequential pattern mining enables you to do that by adding a temporal dimension to the analysis. This technique is applicable, among others, in medicine to calculate the order of a patient’s medical prescriptions and in cybersecurity to predict possible attacks on the system.

Applications of sequential pattern mining include:

  • shopping sequences
  • stock markets
  • natural disasters
  • medical treatments
  • DNA sequencing research

Watch this video if you want to learn more about the points discussed in this article and go deeper into the subject:

Conclusion

Data mining techniques are used to identify patterns in data. They have a wide range of applications in many fields and are increasingly used to develop effective marketing and business development strategies.

Depending on the research objectives and the nature of data, different data mining techniques are applied.

A data mining process is iterative and begins with setting goals, followed by preparing the data, applying various analysis methods, and visualizing the established results.

Unlike machine learning, which uses algorithms to make computers smarter, data mining employs analytical tools to detect patterns. Data mining techniques provide strong data-driven proof and, helping uncover trends and correlations, support decision-making. As such, they are especially effective for business optimization.

Now that you have learned the basics of data mining, you can deepen your knowledge about data processing and analysis.

To do that, check out these articles from our blog:

A Beginner’s Guide to Data Mining Techniques
More from Serokell
Difference between data science, machine learning, and data analytics thumbnailDifference between data science, machine learning, and data analytics thumbnail
deep learning and neural networksdeep learning and neural networks
Classification algorithms; classification in machine learningClassification algorithms; classification in machine learning