That’s a big part of what they do. They got to understand the right algorithms, but just as important you need strong visualization tools so they can see the patters and the charts and the graphics of the trend lines and the heat maps and everything else that a good data mining statistical modeling tool kit, will provide. This is another component included in InetSoft's solution. So at the very highest level those are the core tools that a data scientist needs to do their job and be productive.
What else can I think of in a core curriculum that budding data scientist should have in their toolkit? Well, it's not so much a tool, but it's a frame of reference. They have to understand their business context. They are not building statistical models for fun, for their own fun and amusement. But it's how it work. They need to understand the business context, the business application, what the business is paying them for.
If the business is paying for them to build a churn model to look at causes and variables of customers leaving or customers staying, then they need to understand the products and services the company is providing. The data scientist needs to learn a fair amount about customer churn as a business issue. So they need to acquire an understanding of that. That area is what we called the subject domain.
| #1 Ranking: Read how InetSoft was rated #1 for user adoption in G2's user survey-based index |
|
Read More |
They need to understand the high level business metrics. They are trying to improve an outcome. They are trying to achieve a goal which is improve loyalty, improve retention, improve the rate at which customers accept offers that are made to them. They might reduce the rate at which customers are abandoning shopping carts in the customer portal.
Whatever the outcome is that the business wants to achieve from practical standpoint, the data scientist needs to understand. The data scientist needs be laser focused on that, and that’s the be all and end all. It's not important that neural networks are inherently better in regression, or that logistic regression models are better, or what the underlying algorithm is. There’s many ways to skin the cat of building a model, a model that fits the data.
In the final analysis the data scientist should not develop a fetish for any given statistical modeling approach, and they should have a familiar of the strengths and weaknesses or the applicability of different modeling approaches and modeling tools depending on the business problem.
The last point to make is that you definitely need class room instruction on a topic like this. There’s a significant learning curve. Class structure is important to become a high quality certified data scientist. But hands on experience, as well, is key. It’s like for flying a plane. There are only so many simulations you can do before you have got to get in the cockpit, and you have got to fly a real plane around and hopefully not crash it.
You need hands on laboratory work. You need to be a truly well rounded data scientist and ideally, this is how you’re also getting on the job experience. That will deepen your chops in building good statistical models and tuning them to the data. So keep that in mind: hands-on experience is critically important.