A: Data scientists use a variety of data mining techniques. These approaches use different analytical functions, ask different questions, and use different levels of human input or machine learning algorithms to arrive at decisions. Generally, the most common techniques fall into three main categories:
Descriptive modeling. Descriptive modeling reveals shared similarities in data sets to identify the reasons behind an event or outcome. Some examples of descriptive modeling methods are:
● Clustering – Grouping similar records together to detect anomalies or outliers.
● Association rule learning – Identifying relationships between data points and other records.
● Principal component analysis – Discovering relationships between variables.
● Affinity grouping – Segmenting groups of people with similar goals and interests to analyze behavior.
Predictive modeling. Predictive modeling classifies future events or estimates for unknown outcomes. Some real-world examples of predictive modeling include using someone’s credit score to assess how likely they are to repay a loan or using a person’s past spending behaviors to identify outliers for credit card fraud detection. Examples of predictive modeling methods include:
● Regression – Measuring the strength of relationships between a dependent variable and a series of independent variables.
● Neural networks – Using computer programs and learning algorithms to detect patterns and make predictions.
● Decision trees – Tree-shaped diagrams where each branch represents an event that is likely to occur.
● Support vector machines – Supervised learning models with associated learning algorithms.
Prescriptive Modeling. Prescriptive modeling filters and transforms unstructured data through a process called text mining so that it is ready to be included in predictive models. Prescriptive modeling looks at both internal and external variables to recommend a course of action. Some examples of prescriptive modeling methods include:
● Predictive analytics with rules – Predicting outcomes by developing if/then rules from patterns.
● Marketing optimization – Simulating different types of media in real time to determine the right combination for the highest possible return on investment (ROI).