Machine Learning - eBook (EN)

BUILDING A MODERN STRATEGY FOR ANALYTICS AND MACHINE LEARNING SUCCESS 6 How to implement a modern analytics approach 2. Enable unified data governance and access Every person in your organization who needs access to data should have it—wherever and whenever they need it. This requires secure, seamless access to data lakes and purpose-built data stores and services and being able to access your data no matter where it lives. Data also needs to move seamlessly among applications, systems, and services. For instance, clickstream data from web applications can be collected directly in a data lake, and a portion of that data can be moved to a data ware- house for daily reporting. Or you may want to move sales data from a warehouse into a data lake where it can be stored and analyzed using ML. In tandem with data integration comes the need for centralized data governance and security. Unified data governance involves setting consistent policies across data, services, and applications. It's important to strike the right balance between data security and worker productivity. Consider these steps: Take an enterprise-wide inventory of data resources, who's responsible for them, and who requires access. Create a map of which resources should be made accessible to whom. "One of the benefits of doing this in the cloud is that you can track all these interac- tions," says Pathak. Adopt a zero-trust security approach to access control, which involves limiting access to only those users and applications that require it. Unifying data across silos while tightly controlling access for security reasons requires a delicate data management balance. To harness the insights and innovation that a modern analytics approach enables, consider these four foundational steps. 1. Aggregate data using data lakes An important step in unifying siloed data is to aggregate it into one or more data lakes. The greater the amount of data there is to store, manage, and analyze, the more beneficial it is to create a data lake on a public cloud foundation with infinitely scalable processing resources. Once in the data lake, unified data sets can be consumed again and again, reducing costs and maximizing data value. With this holistic approach, analytics programs account for the whole data picture as they compile and return insights for optimal results. NuData, a Mastercard company, is a testament to the benefits of aggregated, holistic data analytics. It has built fraud detection services that run specialized analytics against very large datasets it stores in Amazon S3 data lakes, powered by ML. NuData runs 26 micros- ervices, each based on a customer use case for detecting a particular type of fraud. While each microservice has its own data lake, it can share data with others. These capabilities help the company correlate petabytes of data each day to identify malicious or erroneous login attempts that might compromise user accounts, explains Justine Fox, NuData's Director of Software Engineering. Unified data governance involves setting consistent policies across data, services, and applications. It's important to strike the right balance between data security and worker productivity.

• Cover