Data Science| season one, lesson 2| processes of data analysis by @nova001 | 10% to steem.skillshare

How is the process of data analysis within Data Science

Data Science and data-driven decision making is an iterative process, not a linear one. However, the standard data analysis cycle usually includes the following steps:

Source

Planning

Determining the objectives and potential outcomes of the project.

Building the Data Model

To build machine learning models, data scientists often use various open source libraries or tools that run on databases. Users often require APIs to make it easier to get data, profile it, visualize it, or develop features. To do this, they need the right tools, as well as access to the right data and other resources such as computing power.

Model Evaluation:

Data scientists need to achieve a high percentage of accuracy for their models. Model evaluation typically generates a complex set of metrics and visualizations to measure the accuracy of models against live data, and to rank them over time to achieve optimal behavior in a production environment. When evaluating models, not only their performance is taken into account, but also the expected underlying behavior.

Explaining Models:

It is not always possible to explain the inner mechanics of the results of machine learning models in human-readable language, but the ability to do so is becoming increasingly important. Data scientists require automated explanations of how the relative weights and importance factors used in forecasting are determined, as well as detailed explanations of the predictions produced by specific models.

Source

Model Deployment:

Applying a trained machine learning model to the right systems is often a complex and time-consuming process. It can be simplified by implementing models as scalable and secure APIs, or by using machine learning models that run in databases.

Model Monitoring:

Unfortunately, it doesn't end with model deployment. Post-deployment monitoring is required to ensure that models work properly. After some time, the data on which the models were trained may cease to be relevant for future forecasts. For example, cybercriminals are constantly implementing new ways to compromise accounts.