What is DSS(Decision Support System) and can you give an example?

[PugoOfficial]

Title says it all. Thanks in Advance!

[Juganog]

decision support system is a computer program that makes it easier to make a decision. A simple example would be a spreadsheet.

 

 

http://en.wikipedia.org/wiki/Decision_support_system

[MikeD]

Decision Support Systems is, as many other fields in computer science, something that you learn in a semester long class and "that's just the basis" according to the teacher!

In fact, we focused our studies in data-driven DSS.

 

Keeping that in mind the summary that I can give you right now is: "A DSS is a management-level computational system that combines data, analytic tools and models to support making semi-structured or non structured decisions." A non structured problem is a new and unusual problem for which no algorithmic solution is known. (Taken from my teacher's book)

 

First, you have to understand that the decision making process goes a little something like: Data --> Information --> Knowledge --> Action

You also have to realize that any DSS must a) present a user friendly interface to the decision makers; b) present data/information in a familiar manner and c) be selective in the presented information (to avoid being bombarded with information. Usualy the IT and analysts will deal with the internals of the system and management will get a pretty report generated by the software! They also need to be as complete in terms of available models since you need to apply many to fully discover patterns in a body of data.

 

DSS employs tools that range from data warehouses to collect and organize data and data mining algorithms to process that data and give you the information you need to make decisions, whatever your business may be.

 

Data warehouses are somewhat like relational databases, except not! They both store information, but that's pretty much the end of similarities.

Relational databases organize data as a bunch of entities that have attributes and the relations between them (hence the name), which is useful in an operational setting (a database of users and logins or books in a library). This model enforces functional dependencies that detect and prevent redundancy and avoiding anomalies when inserting, updating or deleting data and allows you to ask questions such as "What's this users password?" or "How many books in each category published after this date have the word 'X' in the title?".

Data warehouses, on the other hand focus on data analysis rather than on the problems faced by operational systems in the day to day operations of an organization. They follow a multidimensional model that organizes data as a collection of facts (a sale, a deposit/withdraw, the temperature, etc.) associated with a set of attributes organized according to different dimensions, the most important of which being time. This allows the facts to be aggregated at different levels of granularity on one or more of those dimensions (time, location, etc.) by building cubes associated with a given fact whose edges represent dimensions. Then you can execute OLAP queries to ask stuff like "Which car sold most in this region of the country from X to Y day?" or "What was the cost of orthopedic treatments per age group for all years on record at this hospital?".

 

Then we have data mining, a technology with roots in Artificial Intelligence, used to discover hidden patterns and relationships between the stored data and infer rules that predict future behavior.

This process actually starts by fixing some flaws in the input data: missing values must be populated since many companies fail to build warehouses that store all the relevant information and data from different sources must be integrated into the same table to allow mining. Pre-processing also includes removing outliers to reduce noise in the data (using binning techniques) or removing attributes (columns) that are redundant or irrelevant (association rules). Some attributes cannot be processed by mining techniques (text or dates) and need to be transformed or discretized (if there are many possible values, like age).

After all that you can finally apply mining techniques such as dependency derivation (e.g. association rules), clustering (e.g. k-means) or classification (e.g. decision trees, neural networks, bayesian networks).

And then you classify the quality of the model you found by calculating the precision, sensibility, confusion matrices, lift-charts and many more methods.

I skimmed over this last part because it would be even bigger than the first and you can google all these things on your own!

 

This is what a (data driven) decision support system does, more or less! It's a complex system with many parts (data warehouses and mining algorithms being just two) and basically helps you come up with the models that best fit the available data so you can predict what's going to happen in the future.