We provide data-science frameworks to decompose business problems and to re-compose the solutions.
In big data, each data-driven decision-making problem is unique: different combination of goals, constraints, desires and the people concerning each enterprise are different. So, different big data algorithms must be applied depending on each individual problem.
Data scientists tend to set patterns to underlie common business problems. Therefore, data scientists decompose the given business problem into subtasks with the collaboration of the business stakeholders. Then, the subtasks solutions can be composed to solve the overall problem.
In some cases, these subtasks are unique to the particular problem. But, in many others are common data mining tasks.
A problem is unique to a given technology company. However, part of the solution is to “rescue” from historical data of probability a given variable. Once the data have been assembled into a particular format, this estimation fits the mold of one common data-mining task.
διαίρει καὶ βασίλευε (divide and conquer)
There are a large number of data-mining algorithms, but we can decompose them in nine types of tasks that these algorithms address.
Regression (value estimation)
Class probability estimation (classification)
Profiling (behavior description)
The fundamentals are classification, regression, similarity matching and clustering but the rest of them are also common in business. It is needed to introduce important distinctions to be able to decide the best formulation of a given problem.
In data science, a critical skill is to decompose a data-analytics problem into sub-pieces that matches a known task for which tools are available.
Specialists in data science are able to recognize familiar problems and their solutions. It also allows business to avoid spend resources, wasting time and focus attention on other interesting parts of the problem that require human involvement-behaviors.