Predictive analytics uses historical data to predict future events in finance, meteorology, security, economics, insurance, logistics, mobility, and marketing, among others. Generally, historical data is used to create a mathematical model that captures key trends. This predictive model is then applied to current data to predict what will happen next or to suggest actions that can be used to achieve optimal results. Predictive analytics has received a lot of attention in recent years due to major advances in assistive technologies, especially in the areas of big data and machine learning.
Definition of Predictive Analytics
Predictive analytics is a sub-discipline and one of the foundations of so-called “business analytics” in the field of data mining, which deals with predicting future developments. Especially with regard to big data, this method has become indispensable, because it offers a proven technique for analysing large data sets and drawing appropriate conclusions. Data mining plays an important role here, because the information obtained is usually unstructured and must be examined for its usability. The aim is to calculate a probability for the future, whereby predictive analytics is also used to determine trends. The predictors used – a variable in an equation used to predict future behavior – can make fairly accurate predictions about the future. The use of multiple predictors then creates a predictive model that can be used to calculate probable events.
Variants of Predictive Analytics
Predictive models
---
Predictive models (predictive models) use methods from mathematics and computer science to predict an event or outcome. These models predict an outcome in a future state or at a future point in time based on changes in the model inputs. Customers develop the model in an iterative process with a training dataset, then test and validate it to determine the accuracy of its predictions. Different machine learning approaches can be tried to find the most effective model.
The available sample units with known attributes and known performances are called “training patterns”. The units in other samples with known attributes but unknown performances are referred to as “from the [training] sample”. The out-of-sample units are not necessarily chronologically related to the training sample units. For example, the training sample may consist of literary attributes of writings by Victorian authors with known attribution, and the out-of-sample unit may be a newly found font with unknown authorship; a predictive model can help to assign a work to a well-known author. Another example is the analysis of blood spatter at simulated crime scenes, where the unit that is not included in the sample is the actual blood splash pattern from a crime scene. The unit that is not included in the sample can be from the same time as the training sessions, from an earlier time, or from a future time.
Descriptive models
Descriptive Models quantify relationships in data in a way that is often used to classify customers or prospects into groups. Unlike predictive models, which focus on predicting a single customer behavior (e.g., credit risk), descriptive models identify many different relationships between customers or products. Descriptive models don’t assign customers based on their likelihood of taking a particular action. Instead, descriptive models can be used to categorize customers based on their product preferences and stage of life, for example.
Decision models
Decision models describe the relationship between all the elements of a decision—the known data (including the results of predictive models), the decision, and the prediction results of the decision—to predict the outcomes of decisions with many variables. These models can be used for optimization to maximize certain outcomes and minimize others. Decision models are generally used to develop a decision-making logic or set of business rules that generate the desired action for each customer or circumstance. Because of the great complexity of real-world decision problems, there is generally a need for model simplification. One way to simplify this is not to take into account all the characteristics considered possible for the decision-relevant data in the model.
